<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Intento - Medium]]></title>
        <description><![CDATA[Intento enables global companies to translate 20x more on the same budget and makes their customer and employee experience instantly multilingual in 450+ languages through Machine Translation and AI. - Medium]]></description>
        <link>https://medium.com/intento?source=rss----6d3731515c5e---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Intento - Medium</title>
            <link>https://medium.com/intento?source=rss----6d3731515c5e---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 10:59:00 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/intento" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Generative AI for Translation in 2024]]></title>
            <link>https://medium.com/intento/generative-ai-for-translation-in-2024-558d1fe83671?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/558d1fe83671</guid>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[translation]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Konstantin Savenkov]]></dc:creator>
            <pubDate>Mon, 12 Feb 2024 20:57:23 GMT</pubDate>
            <atom:updated>2024-02-12T20:57:23.318Z</atom:updated>
            <content:encoded><![CDATA[<p>In 2023, the landscape of Generative AI (GenAI) for translation saw remarkable advancements, prompting us to conduct an in-depth analysis beyond our annual State of the Machine Translation report. Our study, featuring a rigorous comparison between nine large language models (LLMs) from leading AI developers like Anthropic, Google, and OpenAI, and eight specialized Machine Translation (MT) models, unveils critical insights into the evolving capabilities of GenAI in translation.</p><p>Our main blog has moved, so you may find the full article <a href="https://inten.to/blog/generative-ai-for-translation-in-2024/">https://inten.to/blog/generative-ai-for-translation-in-2024/</a></p><p>Key Findings:</p><ul><li>Performance &amp; Quality: LLMs, while slower, are closing the gap in translation quality compared to specialized MT models, especially in general domain translations. Google’s PaLM 2 models (text-unicorn, text-bison) showed promising results, with text-unicorn ranking highest in general domain translations.</li><li>Domain-Specific Insights: In specialized fields like Legal and Healthcare, traditional MT models still lead, highlighting the tailored efficiency of specialized systems.</li><li>Cost-Efficiency: Generative AI models offer a cost-effective alternative for translation, with Google’s LLMs and LLaMA-2 being 20–25 times cheaper than traditional MT models. However, the slower speed of LLMs is a notable drawback.</li></ul><p>For a detailed dive into our methodology, comparative analysis, and nuanced insights into the strengths and weaknesses of each model, visit <a href="https://inten.to/blog/generative-ai-for-translation-in-2024/">our new blog</a>. Join us as we explore the forefront of machine translation technology, shedding light on the intricacies of generative AI’s role in breaking language barriers.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=558d1fe83671" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/generative-ai-for-translation-in-2024-558d1fe83671">Generative AI for Translation in 2024</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[July 2023: Real-world GenAI applications, event updates, AI news, and more]]></title>
            <link>https://medium.com/intento/july-2023-real-world-genai-applications-event-updates-ai-news-and-more-25cee8cfbe6a?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/25cee8cfbe6a</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[localization]]></category>
            <dc:creator><![CDATA[Intento]]></dc:creator>
            <pubDate>Fri, 11 Aug 2023 12:06:55 GMT</pubDate>
            <atom:updated>2023-08-22T15:14:21.641Z</atom:updated>
            <content:encoded><![CDATA[<p>Understand GenAI’s role in localization, get practical guidance for its implementation, and stay informed about our past and upcoming events 🎤</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z59K0rizaYo2dBz5vUSZoA.jpeg" /></figure><p>Today, the highlights are:</p><p>🌐 Real-world GenAI applications for enhanced localization programs</p><p>💡Getting started with GenAI in localization: Expert advice and practical tips</p><p>🚀 Oracle Cloud now available directly in Intento</p><p>🍽️ Intento &amp; e2f dinner: A night of innovation and insight</p><p>💎 Intento: Proud Diamond sponsor for the Women in Localization 15th Anniversary Gala Party</p><p>🤖 Anthropic’s second-gen AI chatbot Claude 2 released</p><p>…and more!</p><h3>Real-world GenAI applications for enhanced localization programs</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EIjQP1g1h_m35daYcU3qUA.png" /></figure><p>We’ve sifted through our latest webinar, unearthing crucial perspectives from industry experts on GenAI’s role in localization.</p><p>💡 Intento CEO Konstantin Savenkov introduced “zooming out” as a strategic approach to new technology</p><p>🔍 Olga Hergül from Şişecam and Bodo Vahldieck from VMware discussed leveraging large language models for quality improvement and process streamlining</p><p>🚀 Hartmut von Berg from GoTo is integrating GPT into marketing localization</p><p>🌍 Tommaso Rossi from Expedia sees GPT’s potential in localizing content in over 20 languages</p><p>Find more on that <a href="https://blog.inten.to/real-world-genai-applications-global-enterprises-using-ai-for-enhanced-localization-programs-93b05e3387a6">in our blog</a> ✨</p><h3>Getting started with GenAI in localization: Expert advice and practical tips</h3><p>In the second panel of our webinar, Beatriz Verdasco from Trendyol revealed how GenAI can transform e-commerce localization. Ray Flournoy, formerly with Etsy, shared strategies for enhancing AI-generated translations. Adam Ali from the Government of Canada and Peter Tatishchev from Disney+ tackled data privacy, security, and GPT model integration. <a href="https://blog.inten.to/getting-started-with-genai-in-localization-expert-advice-and-practical-tips-34d1d8551dea">Check out the full summary</a> 👀</p><h3>Oracle Cloud now available directly in Intento</h3><p>You can now directly use the <a href="https://inten.to/api-platform/api/oracletranslate">Oracle Cloud</a> translation service in Intento, along with dozens of other translation services, without the need to sign a separate vendor contract.</p><h3>Intento &amp; e2f dinner: A night of innovation and insight</h3><p>In early July, we hosted a dinner with <a href="https://www.e2f.com/">e2f</a>. It was great to see familiar faces, share summer plans and… delve into the exciting potential of GenAl. We discussed everything from real-time and culturally-aware translations to major AI challenges like biases, depersonalization, security concerns, and the build vs buy conundrum. What a night! ✨</p><h3>Intento: Proud Diamond sponsor for the Women in Localization 15th Anniversary Gala Party</h3><p>We’re privileged to back <a href="https://womeninlocalization.com/">Women in Localization</a> and their mission, a cause close to our hearts. This year, we’ve stepped up as the Diamond sponsor for their grand Gala Party! <a href="https://womeninlocalization.com/15anniversary/">Get your ticket</a> and join us on October 12th at The Glasshouse in San Jose. It’s a chance to celebrate 15 years of industry success and network with the leading minds and visionaries in the field 🎩</p><h3>Anthropic’s second-gen AI chatbot Claude 2 released</h3><p>Shifting to AI news — Anthropic has launched <a href="https://www.anthropic.com/index/claude-2">Claude 2</a>, the successor to its initial commercial AI text-generating model, now in beta across the US and UK. This enhanced model demonstrates superior performance over Claude 1.3, showing improvements in areas like bar exam and coding tests. Trained on more up-to-date data and having a wider context window, Claude 2 can generate and analyze more text.</p><h3>News on AI and MT developments</h3><ul><li><a href="https://techcrunch.com/2023/07/12/lince-llm/">Clibrain joins the generative AI race with Lince, an LLM optimized for Spanish</a></li><li><a href="https://arxiv.org/abs/2307.06018">PolyLM: An Open Source Polyglot Large Language Model</a></li><li><a href="https://finance.yahoo.com/news/machine-translation-market-revenue-cross-090000217.html">Machine Translation Market revenue to cross USD 5 Billion by 2035</a></li><li><a href="https://www.eurekalert.org/news-releases/994611">A neural machine translation system for all the Romance languages of the Iberian Peninsula</a></li><li><a href="https://restofworld.org/2023/3-minutes-with-asmelash-teka-hadgu/">The AI startup outperforming Google Translate in Ethiopian languages</a></li><li><a href="https://news.slashdot.org/story/23/07/31/0249243/forget-subtitles-youtube-now-dubs-some-videos-with-ai-generated-voices">YouTube now dubs some videos with AI-generated voices</a></li><li><a href="https://www.cio.com/article/648558/nasa-ibm-team-up-to-build-llms-that-can-help-fight-climate-change.html">NASA, IBM team up to build LLM that can help fight climate change</a></li><li><a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-in-2023-generative-ais-breakout-year">The state of AI in 2023: Generative AI’s breakout year</a></li><li><a href="https://www.cnbc.com/2023/08/03/alibaba-launches-open-sourced-ai-model-in-challenge-to-meta.html">Chinese tech giant Alibaba challenges Meta with open-sourced AI model launch</a></li><li><a href="https://www.computerweekly.com/news/366545937/ServiceNow-Q2-2023-23-year-on-year-growth-with-GenAI-to-fore">ServiceNow Q2 2023: 23% year-on-year growth with GenAI to fore</a></li><li><a href="https://erp.today/capgeminis-research-finds-executives-believe-genai-benefits-will-outweigh-the-concerns/">Capgemini’s research finds executives believe genAI’s benefits will outweigh the concerns</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=25cee8cfbe6a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/july-2023-real-world-genai-applications-event-updates-ai-news-and-more-25cee8cfbe6a">July 2023: Real-world GenAI applications, event updates, AI news, and more</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Getting started with GenAI in localization: Expert advice and practical tips]]></title>
            <link>https://medium.com/intento/getting-started-with-genai-in-localization-expert-advice-and-practical-tips-34d1d8551dea?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/34d1d8551dea</guid>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[localization]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-translation]]></category>
            <dc:creator><![CDATA[Intento]]></dc:creator>
            <pubDate>Mon, 17 Jul 2023 22:26:40 GMT</pubDate>
            <atom:updated>2023-07-17T22:26:40.053Z</atom:updated>
            <content:encoded><![CDATA[<h4>Discover how to effectively and responsibly use generative AI in localization with insightful highlights from experts at Disney+, the Government of Canada, and Trendyol.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3Bs2XxElyuW7Mvst99OzTA.jpeg" /></figure><p>Intento’s recent webinar, “GPT in Localization: When the Tide Goes Out,” featured two panels discussing the application of GenAI in localization. The second panel specifically addressed practical applications, data privacy, and security. Specialists from Trendyol, Government of Canada, Disney+ , and independent experts provided essential guidance on getting started, overcoming implementation challenges, and utilizing generative pre-trained transformers to improve localization workflows.</p><p>Learn from their experiences with content generation, translation, quality control, accessibility, workforce development, and large language model (LLM) integration. You can also watch the full recording of the webinar <a href="https://www.youtube.com/watch?v=xSrmo5uxyls&amp;t=471s">here</a>.</p><h3>Using GenAI in e-commerce for content generation and source quality improvement</h3><p><strong>Beatriz Verdasco, Head of Localization at </strong><a href="https://www.trendyol.com/"><strong>Trendyol</strong></a>, believes that GenAI can be a game-changer for e-commerce localization. However, it is crucial to identify appropriate use cases, manage volume and speed issues, and guarantee high-quality results to get the maximum returns:</p><ul><li>Focus on scenarios where machine translation is not feasible due to high volume, speed, or quality concerns, and consider content generation with GenAI as a cost-efficient alternative;</li><li>Experiment with GenAI in source quality improvement, particularly for lower-resource languages and challenging language pairs like Turkish to German;</li><li>Address AI hallucinations, like generating inappropriate suggestions or made-up details, by refining the prompts or adjusting the model to avoid including unwanted information;</li><li>Remember that using English as the source language can yield better results with GenAI due to its potential limitations in processing prompts in other languages.</li></ul><h3>Enhancing AI-generated translations with strategies for quality control and user experience</h3><p><strong>Ray Flournoy, Former VP of Product at </strong><a href="http://etsy.com/"><strong>Etsy</strong></a>, gives several tips on utilizing GenAI to generate precise translations, underscoring the importance of ongoing enhancement, feedback mechanisms, accessible tools, and multi-input methods:</p><ul><li>Tackle the issue of inaccuracies and distortions in AI-generated content by employing rigorous quality assessment methods and error identification strategies;</li><li>Establish a communication cycle between translators and AI models, enabling the AI to improve performance by learning from revised translations;</li><li>Investigate the capabilities of GenAI to develop an efficient human-in-the-loop system that functions intelligently and reliably and fosters seamless interaction for enhanced translation;</li><li>Build dependable, easy-to-use translation solutions for sellers of diverse linguistic and technological skills, enabling them to access international markets;</li><li>Explore the potential of GenAI to take in various inputs, including original listing descriptions and images, to enhance the translation process.</li></ul><h3>Addressing data privacy, security, and workforce preparation in government settings</h3><p><strong>Adam Ali, IT Strategist at the </strong><a href="https://www.canada.ca/"><strong>Government of Canada</strong></a><strong>, </strong>emphasizes the importance of data privacy, security, and workforce adaptation in AI-driven localization, focusing on policy development, continuous learning, and secure AI models:</p><ul><li>Develop and implement regulatory frameworks, usage guidelines, and data handling policies for AI-generated content and GPT models in collaboration with legal and IT departments;</li><li>Build custom large language models with safeguards to reduce data leak and bias risks, ensuring regulatory compliance;</li><li>Foster a learning culture for translators by offering AI technology training, prompt engineering, and refining organizational procedures for robust AI governance and policy structures;</li><li>Employ ChatGPT for content improvement, translation support, translation memory enrichment, and machine translation accuracy evaluation while ensuring security and privacy through data encryption.</li></ul><h3>Optimizing GPT model integration and security in content platforms</h3><p><strong>Peter Tatishchev, Internationalization Product Manager at </strong><a href="https://www.disneyplus.com/"><strong>Disney+</strong></a>, highlights the potential of LLM models for streamlining localization workflows, enhancing translation quality, addressing security concerns, ensuring vendor accountability, and optimizing in-house model training:</p><ul><li>Enable smooth incorporation of LLM models into localization workflows through essential tooling support and cooperation between engineering and localization teams;</li><li>Improve the creation of target texts using the same inputs as the source text rather than translating from English for better context and translation quality;</li><li>Consider developing in-house language models, isolated from the broader internet, to guarantee content security;</li><li>Ensure vendor responsibility for data protection when employing third-party solutions that may use machine translation or content creation services;</li><li>Explore the combination of public and proprietary data to train in-house models to ensure sufficient training data while maintaining control over the model.</li></ul><h3>Conclusion and takeaways</h3><p>Effective implementation of GenAI in localization offers numerous benefits. However, localization experts should define appropriate use cases, establish quality management, prioritize data protection, improve user experience, and streamline model incorporation.</p><p>Through continuous improvement, they can utilize GenAI’s comprehensive features, providing reliable, secure, and sophisticated solutions that empower them to tackle complex linguistic issues and adapt to the changing localization landscape.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=34d1d8551dea" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/getting-started-with-genai-in-localization-expert-advice-and-practical-tips-34d1d8551dea">Getting started with GenAI in localization: Expert advice and practical tips</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Real-world GenAI applications: Global enterprises using AI for enhanced localization programs]]></title>
            <link>https://medium.com/intento/real-world-genai-applications-global-enterprises-using-ai-for-enhanced-localization-programs-93b05e3387a6?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/93b05e3387a6</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[localization]]></category>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[machine-translation]]></category>
            <dc:creator><![CDATA[Intento]]></dc:creator>
            <pubDate>Wed, 28 Jun 2023 15:27:05 GMT</pubDate>
            <atom:updated>2023-06-28T20:01:52.595Z</atom:updated>
            <content:encoded><![CDATA[<h4>Discover insights from four industry experts on the challenges and benefits of applying generative AI to localization — with real-life examples for optimal results across different sectors and languages.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8_-TH5lav1P47oYWS95JwA.jpeg" /></figure><p>The localization industry is rapidly evolving, and GenAI is a big reason why. Experts from Expedia, GoTo, Şişecam, and VMware shared their experiences with AI in localization workflows at our webinar “<a href="https://www.youtube.com/watch?v=xSrmo5uxyls&amp;t=471s">GPT in Localization: When the tide goes out…</a>”, spotlighting the most significant results.</p><p>For those who missed the webinar, this summary provides key takeaways from the panelists’ strategic insights. The full recording of the webinar is available <a href="https://www.youtube.com/watch?v=xSrmo5uxyls&amp;t=471s">here</a>.</p><h3>Shifting focus to high-value localization tasks with GenAI</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*kHaZ3l5Z3CAAtgki" /><figcaption>“A path to a bigger zoom out” in localization</figcaption></figure><p>In his keynote, Konstantin Savenkov, Intento’s co-founder and CEO, highlighted the massive impact of large language models on the language industry and introduced the concept of “zooming out” as a strategic method for adapting to innovative technology and maintaining a competitive edge.</p><blockquote>“Zooming out is the best approach to new technology. You have more time on the more important things, but keep the original task in your scope, as it costs you close to nothing, thanks to automation. It gives you an upper hand in any competition, be it a business competition or an internal competition,” Konstantin explained, emphasizing that the emergence of large language models is a perfect case for applying this zooming out technique.</blockquote><p>As we shared on our previous webinar in December 2022, even before GPT-4, advanced automation in localization expanded beyond Machine Translation to include Source Quality Improvement, Automatic Post-Editing, and Translation Quality Estimation. These pillars empower localization teams to address diverse translation use cases.</p><p>Today, technology advances further, providing a path to a bigger zoom-out. With translation sufficiently automated, localization leaders can look at a bigger scope. Creating content based on inputs, context, and style guides. Making sure the content created across the company fits all the requirements. Translating the content, checking that it aligns with guidelines. Presenting it to users in a culture and identity-aware, personalized fashion. Analyzing user experience, or “language experience”, collecting feedback, and then incorporating it into the process.</p><p>Gain insight into the challenges, benefits, and real-life examples of applying generative AI to localization across various sectors and languages, as shared first-hand by industry experts from top global companies.</p><h3>Overcoming complex syntax and maintaining brand voice with GenAI at Şişecam</h3><p><a href="https://www.sisecam.com.tr/">Şişecam</a>, a leading Turkish global glass manufacturer, operating in 14 countries and selling in 150+ countries. Olga Hergül, Global Content and Localization Program Manager at Şişecam, shared that they maintaining a consistent brand voice for its linguistically diverse audience is challenging due to the intricate Turkish syntax. Recently, she has been exploring the potential of large language models like OpenAI’s GPT.</p><p>Although they haven’t yet deployed any workflows involving large language models into production, Şişecam has been testing the technology for two primary use cases: Source Quality Improvement and Automatic Post-Editing.</p><blockquote>“With generative AI, we tested different process combinations, such as source quality improvement for Turkish plus machine translation, or machine translation plus automated post-editing for English, or all three stages in a row,” Olga Hergül, localization program manager at Şişecam, elaborates. “The initial results have been quite promising, with AI working surprisingly well for Turkish in terms of source quality improvement.”</blockquote><p>Yet, there’s a catch. Although GPT performed well for corporate communications, such as internal regulations and employee communications, requiring minimal or no human intervention in some cases, it sometimes can alter the tone of voice and style, particularly in content like press releases or interviews.</p><p>That’s why Olga emphasized the need to adjust style guides for large language models. This way, Şişecam’s content can stay true to its brand identity and still connect with diverse audiences.</p><p>Summing up her observations, Olga said that Şişecam is eager to further test and implement GPT or alternative large language models into localization workflows.</p><h3>Achieving cost-effective, high-quality content with GPT and Whisper at VMware</h3><p>Bodo Vahldieck, Senior Globalization Quality Manager at <a href="https://www.vmware.com/">VMware</a>, described how they explore the potential of GenAI tools like GPT and Whisper to enhance VMware localization program. They aim to improve source content, streamline processes, and speed up reference process management while maintaining top-notch content quality standards and integrating AI with current workflows.</p><p>By adopting AI, VMware aims to save costs, speed up turnaround times, and maintain consistent quality. Bodo recognizes AI’s potential to improve linguistic quality, eliminate unnecessary words, and enhance speech recognition accuracy.</p><blockquote>“With AI, we see now for the first time we can give the authors a tool to check errors in the source text,” Bodo Vahldieck, Senior Globalization Quality Manager, states, emphasizing the importance of improving source text, involving authors in the process, and making the process more visible</blockquote><p>They are exploring the automation of source content quality improvement and considering implementing AI-powered translation services in production. Also, VMware is evaluating OpenAI’s Whisper technology to advance speech recognition and refine spoken language in video subtitles. In their current workflows, human reviewers play a crucial role in post-translation stages, detecting and addressing quality concerns.</p><p>VMware’s approach to creating cost-effective, high-quality content is a blend of teamwork, AI innovation, and the right balance of AI technologies with human expertise. By further enhancing its processes with cutting-edge AI technology and data analysis, VMware seeks to advance its marketing localization efforts, deliver top-quality results, and improve turnaround times.</p><h3>Streamlining GoTo’s marketing localization with GenAI</h3><p>Hartmut von Berg, Senior Director of Localization at <a href="https://www.goto.com/">GoTo</a>, a company known for its conferencing and IT support products, has been exploring the potential of GenAI technology for marketing localization. Following its rebranding from LogMeIn, the company is now integrating AI technology into its marketing localization strategies. This technology has been tested in a beta product to create compelling customer campaigns. Their prior AI experience mainly focused on translating technical documents, prioritizing information communication, and addressing user concerns over fluency and accuracy.</p><p>GoTo prioritizes adapting content without compromising brand guidelines for enhanced quality assurance and market adaptation. Hartmut von Berg, Senior Director of Localization at the company, shares their strategy for establishing consistent brand localization guidelines and implementing powerful GenAI for quality assurance.</p><p>GoTo is exploring running their own GPT instance and applying an intentional bias to achieve accurate and personalized results without external biases from large corporations like Apple or Microsoft. In collaboration with Intento, they aim to assess the potential of AI in marketing localization and identify the most effective method for incorporating it into their workflow.</p><p>However, they must overcome challenges such as altering style guides to match GPT comprehension, handling repetitive terminology that defines their brand voice, and navigating country and culture-specific biases.</p><blockquote>“We believe now we need to rewrite [the brand guidelines] and tailor them for the machine to get the best results out of it,” Hartmut states.</blockquote><p>He also noted that breaking down large guidelines into smaller, meaningful blocks can enhance AI’s attention and focus on specific aspects, while finding that inclusivity and formality are more easily detected, and other aspects may necessitate further fine-tuning.</p><p>Despite these challenges, GoTo’s mix of human expertise and AI is designed to deliver an engaging and culturally relevant user experience. This approach streamlines the localization process and unleashes the full potential of AI-assisted localization, helping GoTo maintain a consistent brand identity across various markets.</p><h3>Expedia’s transliteration and dialect conversion improvements using GenAI</h3><p>Tommaso Rossi, <a href="https://www.expediagroup.com/">Expedia</a>’s Head of Localization, discussed using early access to GPT to enhance their global travel platform. After integrating the AI-powered connector for personalized trip suggestions, Expedia is now exploring GPT’s potential for localization in over 20 languages.</p><p>Initial tests for multilingual content creation showed that the technology is not yet ready for off-the-shelf use, requiring additional human assistance in fact-checking and aligning with industry-specific terminology. Rossi highlighted the significance of prompt engineering in improving GPT’s output quality and determined that GPT is comparable to their current machine translation providers.</p><blockquote>Rossi also noted that the quality produced by GPT is very much comparable to the machine translation engines they’re currently using, adding that “this means that potentially with further testing, we could discover that with one solution, which is the GPT technology, we could potentially reach the same level [of quality] that we get by [using] the right [MT] engine for the right language pair and the language content type across the board.”</blockquote><p>Rossi also said GPT stands out in transliteration and dialect conversion, making it a promising field compared to traditional machine translation methods. Initial tests on geography content and dialect conversions, like switching from US English to UK English, showed promising results with minimal errors and high accuracy. Plus, GPT has proven its success in transliteration for languages like Japanese, Chinese, Korean, Greek, and Hebrew. Expedia is keen on further testing and exploring advanced methods for different types of content.</p><p>His team now plans to refine their testing and implementation strategies to improve localization, content creation, and user experience. As they continue exploring GPT’s potential, users worldwide can benefit from enhancements in Expedia’s multilingual services.</p><h3>Key takeaways and prospects</h3><p>Generative AI tools, including large language models and automation, can help localization professionals focus on strategic tasks and streamline workflows while effectively addressing complex syntax, brand voice consistency, content quality, and customization for target markets.</p><p>Using GenAI leads to faster turnarounds, better results, and reduced human effort across different languages, industries, and contexts. The future of localization relies on continuous learning, adaptation, and experimentation with this innovative technology.</p><p>As we move forward, it is essential for industry players to stay informed, engage with emerging technologies, and adapt their strategies accordingly. In doing so, they can drive innovation, ensuring that products and services resonate deeply with diverse audiences worldwide, ultimately contributing to the growth and success of their organizations.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=93b05e3387a6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/real-world-genai-applications-global-enterprises-using-ai-for-enhanced-localization-programs-93b05e3387a6">Real-world GenAI applications: Global enterprises using AI for enhanced localization programs</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[June 2023: The State of Machine Translation 2023, new features, AI updates, and more]]></title>
            <link>https://medium.com/intento/june-2023-the-state-of-machine-translation-2023-new-features-ai-updates-and-more-2eab3b425558?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/2eab3b425558</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[localization]]></category>
            <dc:creator><![CDATA[Intento]]></dc:creator>
            <pubDate>Tue, 20 Jun 2023 21:14:33 GMT</pubDate>
            <atom:updated>2023-06-21T15:25:32.268Z</atom:updated>
            <content:encoded><![CDATA[<h4>Stay ahead in the industry with insights from Intento’s 2023 State of Machine Translation report, MT evaluation techniques, recent event updates, much-anticipated features, and AI developments from the past month 🚀</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HN_ajyc8rph56BhtrkviMg.jpeg" /></figure><p>In this edition, we’re covering:</p><p>📊 Intento’s 2023 State of Machine Translation report</p><p>📖 New course at Intento Machine Translation University: Automated scoring and evaluation of MT Engines</p><p>🚀 Batch uploads in the Intento Translation Portal</p><p>⚡ Intento Translator for Verint</p><p>🌟 SAP Document Translation, Mirai Translator, and RoyalFlush Translate</p><p>🎉 Intento Translator for Chrome</p><p>🌊 Intento webinar: “GPT in Localization: When the tide goes out…”</p><p>💬 Intento presentation at LocWorld49 (with IKEA, Qualtrics, and Playrix): “Multilingual Customer and Employee Experience: Everything, Everywhere, All at Once”</p><p>…and more!</p><h3>The State of Machine Translation 2023 is here 🔥</h3><p>The 2023 edition of our State of Machine Translation report (our 7th annual) is packed with data-driven insights on how 37 MT Engines (including 5 LLMs!) perform across 22 language pairs in 9 content domains. <a href="https://inten.to/machine-translation-report-2023/">Get your free copy now</a> ⬅️</p><p>In this collaborative edition with <a href="https://www.e2f.com/">e2f</a>, we harnessed the COMET metric to identify top models, analyzed pricing and language support, and offered domain-specific recommendations. Plus, for the first time, we’ve evaluated models from AI21, Anthropic, OpenAI, and NVIDIA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jNCTJVBhPqM8zrhsFcQTVQ.jpeg" /></figure><h3>Master your MT evaluation expertise 🎓</h3><p>Explore our new Machine Translation University chapter on <a href="https://inten.to/machine-translation-university/automated-scoring-and-evaluation-of-mt-engines/">automated scoring and evaluation of MT Engines</a> 📖 Learn how to set evaluation goals 🎯, effectively combine automatic and human assessments, make the most of scoring, and refine your sampling strategy. A must-read on how to turn your MT evaluation into a data-driven, cost-saving approach 💡</p><h3>Batch uploads in Intento Translation Portal</h3><p>You can now upload up to 10 files at once when using our <a href="https://inten.to/translation-portal/">Translation Portal</a> — including PDFs. The Translation Portal is part of the MT Hub for Employee Experience subscription.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*E3cH47UTW7c0lEyWZ3MbyQ.jpeg" /></figure><h3>Improved Intento Translator for Verint Community</h3><p><a href="https://community.telligent.com/marketplace/mp/listings/intento-translator-for-telligent">Intento Translator for Verint Community</a> just got a brand new web translation algorithm, resulting in performance improvements and increased flexibility in visual and functional configuration. The Intento Translator for Verint Community is part of the MT Hub for Employee Experience subscription.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1021/1*4v6SIl35pVoWcvotoknZJw.jpeg" /></figure><h3>Three new MT providers added to Intento</h3><p>We’ve added <a href="https://help.sap.com/docs/SAP_TRANSLATION_HUB/9f73362817cd48339dd8a6acba160f7f/a2dedd7861624a1a82d7ec7ea431a8e4.html?locale=en-US">SAP Document Translation</a> (another SAP engine), <a href="https://miraitranslate.com/en/">Mirai Translator</a>, and <a href="https://translate.10jqka.com.cn/">RoyalFlush Translate</a> to our platform. Use your Intento credentials for direct access.</p><h3>Enhanced Intento Translator for Chrome</h3><p><a href="https://chrome.google.com/webstore/detail/intento-translator/ofmfoeohdhdgcakhdcegeiljlgammpjf">Intento Translator for Chrome</a> now runs an upgraded web translation engine, making it smarter and faster than ever. Translate any content on any web page using a simple keyboard shortcut.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*pUgQr57_WaqBswBBHuhLPg.gif" /></figure><h3>Intento webinar: “GPT in Localization: When the tide goes out…”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1018/1*7go94m1oaqOxgL-kmERljg.png" /></figure><p>In our recent <a href="https://inten.to/webinars/gpt-in-localization/">webinar dedicated to real-world GenAI applications</a>, eight industry experts from different sectors (GoTo Technologies, Şişecam, Expedia Group, Disney, Etsy, Government of Canada, VMware, and Trendyol) discussed the powerful impact of Generative AI on localization and content creation, shared their first-hand experiences, and offered practical advice for using it in your workflows.</p><p>The webinar also includes a mini-session on the history of Large Language Models by Intento CTO and co-founder Grigory Sapunov — and is essential for anyone looking to drive innovation in their localization program.</p><p><a href="https://www.youtube.com/watch?v=xSrmo5uxyls">Watch the full recording</a> 👀</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x-MjlG__lZwE96dJW78HVQ.png" /><figcaption>“A path to a bigger zoom out” in localization</figcaption></figure><h3>Intento presentation at LocWorld49: “Multilingual Customer and Employee Experience: Everything, Everywhere, All at Once”</h3><p>At this year’s <a href="https://locworld.com/locworld49-malmo-conference-program/">LocWorld conference</a>, Intento CEO Konstantin Savenkov hosted a panel featuring Johan Sporre from IKEA, Giulia Tarditi from Qualtrics, and Antonina Vasyukova from Playrix. The discussion centered on the benefits of real-time machine translation in enhancing language accessibility for both customer and employee experiences. Stay tuned for a summary with the key insights 👀</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*efOxdxvPbsIF28tiHDjj_g.jpeg" /></figure><h3>Google launches PALM 2 to counter GPT-4</h3><p>Google’s next-gen model, <a href="https://blog.google/technology/ai/google-palm-2-ai-large-language-model/?utm_campaign=Social%20Media&amp;utm_content=248531755&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">PaLM 2</a>, established a new standard in translation accuracy and regional dialect handling, featuring improved multilingual capabilities and faster deployment. Get ready for Google’s forthcoming, versatile Gemini model as well.</p><h3>LIMA: Less Is More for Alignment</h3><p><a href="https://www.meta.com/">Meta’</a>s new 65B LLaMa language model, <a href="https://huggingface.co/papers/2305.11206">LIMA</a>, has demonstrated solid performance, excelling in specific response formats and handling complex queries. Outperforming GPT-4, Bard, and DaVinci003, LIMA presents a significant advancement in AI 💫</p><h3>SelFee, a self-revising LLM</h3><p><a href="https://kaistai.github.io/SelFee/?utm_campaign=Social%20Media&amp;utm_content=251925879&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">SelFee</a> is a self-revising LLM by <a href="https://lklab.kaist.ac.kr/">LK LAB</a> empowered by its own feedback generation. With self-feedback on responses and continuous improvement through revisions, SelFee’s 7B and 13B models excel in creative writing and long-form text generation. They outperform LLaMA, Alpaca, Vicuna, and Guanaco and are comparable to ChatGPT.</p><h3>News on AI and MT developments</h3><ul><li><a href="https://deepgram.com/state-of-voice-technology-2023/">The State of Voice Technology in 2023</a></li><li><a href="https://venturebeat.com/ai/openai-is-rolling-out-chatgpt-plugins-to-plus-subscribers-heres-why-thats-a-big-deal/">OpenAI is rolling out ChatGPT Plugins to Plus subscribers</a></li><li><a href="https://arxiv.org/abs/2305.16307">IndicTrans2: Enhancing Machine Translation for 22 Scheduled Indian Languages</a></li><li><a href="https://www.researchgate.net/publication/371124206_Do_GPTs_Produce_Less_Literal_Translations">Do GPTs Produce Less Literal Translations?</a></li><li><a href="https://www.trustedreviews.com/how-to/how-to-delete-all-your-data-from-chatgpt-4331607">How to delete all your data from ChatGPT</a></li><li><a href="https://www.wired.com/story/fast-forward-gpt-4-minecraft-chatgpt/">Minecraft bot “Voyager” uses GPT-4 for in-game actions</a></li><li><a href="https://www.theverge.com/23739382/discord-memrise-membot-language-learning-ai">Memrise launches GPT-powered language-learning chatbot</a></li><li><a href="https://slator.com/30-new-jobs-language-industry-ceos-expect-to-hire-ai-age/">New AI language roles at LSPs: Prompt Debugger, AI Ethicist, Post-editor of LLM MT Output</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2eab3b425558" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/june-2023-the-state-of-machine-translation-2023-new-features-ai-updates-and-more-2eab3b425558">June 2023: The State of Machine Translation 2023, new features, AI updates, and more</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[May 2023: Machine Translation University updates, OCR in the MT Portal, new languages, and more]]></title>
            <link>https://medium.com/intento/may-2023-mt-university-updates-ocr-in-the-mt-portal-new-languages-and-more-f5a09900e3be?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/f5a09900e3be</guid>
            <category><![CDATA[localization]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[digest]]></category>
            <dc:creator><![CDATA[Intento]]></dc:creator>
            <pubDate>Tue, 16 May 2023 03:10:20 GMT</pubDate>
            <atom:updated>2023-05-16T18:57:24.104Z</atom:updated>
            <content:encoded><![CDATA[<h4>New Machine Translation University chapter on selecting MT systems, new languages available through MT Hub, and a handful of AI news that made April so remarkable ✨</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bNp7eDxkbABekOT4rfHUiw.jpeg" /></figure><p>April unveiled 🌸:</p><ul><li>New chapter 📖: Selecting MT systems based on available assets</li><li>New languages from Microsoft and ModernMT</li><li>PDF translations (with OCR!) now available via Intento Portal</li><li>Glossary labels — New!</li><li>Intento Translator for Chrome — Updated!</li><li>RoyalFlush Translate—New!</li><li>Dolly 2.0 by Databricks</li><li>HuggingChat by Hugging Face</li><li>Building LLM apps for production</li><li>Eventful April</li></ul><h4>New Machine Translation University chapter 📖: Selecting MT systems based on available assets</h4><p>Where do you start when choosing MT systems? 🤔 Check out <a href="https://inten.to/machine-translation-university/selecting-mt-systems-based-on-available-assets/">the latest addition to our Machine Translation University</a> covering:</p><p>🌍 Language support<br>💰 Pricing<br>📚 Glossary availability<br>🧠 TM size<br>🎯 Domain adaptation<br>🔒 Compliance &amp; security</p><p><a href="https://inten.to/machine-translation-university/selecting-mt-systems-based-on-available-assets/">Dive in</a> and remember — in MT, cost and quality don’t always correlate 🤷🏻‍♀</p><h4>New languages from Microsoft and ModernMT</h4><p>Microsoft Translator recently added <a href="https://www.microsoft.com/en-us/translator/blog/2023/04/18/announcing-four-new-languages-konkani-maithili-sindhi-and-sinhala/">Konkani, Maithili, and Sindhi</a> from India 🇮🇳, <a href="https://www.microsoft.com/en-us/translator/blog/2023/04/18/announcing-four-new-languages-konkani-maithili-sindhi-and-sinhala/">Sinhala</a> from Sri Lanka 🇱🇰, and <a href="https://www.microsoft.com/en-us/translator/blog/2023/04/20/introducing-lower-sorbian/">Lower Sorbian</a> from Germany 🇩🇪 — all ready for use via Intento MT Hub!</p><p>And ModernMT <a href="https://multilingual.com/translated-unveils-adaptive-machine-translation-service-in-200-languages/">now boasts 200 languages</a>, with 29 new for our platform.</p><p>Our <a href="https://inten.to/demo/">live demo</a> is a great place to try how different vendors stack up for other language pairs 👉🏻 <a href="https://inten.to/api-platform/ai/text/translate">https://inten.to/api-platform/ai/text/translate</a>.</p><h4>PDF and OCR translations now available via Intento Portal</h4><p>Intento MT Portal now supports the translation of PDF files. It even supports scanned PDFs using the embedded OCR feature. The MT Portal is available to customers who have a subscription to <a href="https://inten.to/employee-experience/">MT Hub for Employee Experience</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/860/1*eeefZy9oO0wePxtgJzQ98g.png" /></figure><h4>Glossary labels — new feature alert 🔔</h4><p>You can now add labels to your glossaries for easier navigation. Did you know Intento Glossaries work in a provider-agnostic fashion and can be case-sensitive? <a href="https://help.inten.to/hc/en-us/articles/4486733211804-Intento-Glossaries">More on that here</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Vid_dHLLKsX0DvnoVNiOmw.png" /></figure><h4>Eventful April</h4><p>In late April, Pavel Doronin (our Product Director) presented on <a href="https://xtmlive.com/agenda/">XTM Live</a> how we <strong>augment MT with Automatic Post-Editing</strong> to apply glossaries, and control tone of voice, gender, and other aspects of translations. Of course, GPT is involved!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DHSXPM64rFzVVSvDCyzULA.jpeg" /></figure><p>At the same conference, Andy Nikulin (Customer Success Director) shared our case study on gendered translations at scale, discussing how Intento tried out GPT for adjusting gender and tone of voice in video translations (case study coming soon!)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*3q3ZC9Z3qS6BD9yV" /></figure><p>Earlier, at the end of March, we sat down with Edith from NetApp and Theresa from Nike to talk about <a href="https://www.eventbrite.com/e/wlsp-beyond-mt-navigating-the-new-ai-driven-localization-landscape-registration-580912273247">what lies beyond MT</a> (recording available) and how to navigate today’s increasingly AI-driven localization landscape.</p><p>Finally, at the GALA conference in Dublin, IKEA shared their first successful experiments with <strong>GPT-based Automatic Post-Editing</strong> — a 25% edit distance reduction on top of their custom NMT.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TeWWqhtYp9QulTAQ" /></figure><h4>RoyalFlush Translate</h4><p>A new addition to our lineup of MT providers, RoyalFlush Translate excels in seamless global financial communication. Drawing on extensive financial data and comprehensive knowledge, it provides high-quality translations for text, documents, speech, and images in 10+ languages, including Malay, Vietnamese, and Indonesian.</p><h4>Dolly 2.0 by Databricks</h4><p>In mid-April, <a href="https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm?utm_campaign=Social%20Media&amp;utm_content=245190681&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">DataBricks released Dolly 2.0</a> — an open-source, instruction-tuned 12B-parameter LLM based on the EleutherAI’s Pythia-6.9b model and fine-tuned on a high-quality human-generated dataset.</p><p>Dolly 2.0 is available for commercial use without any API fees or third-party risks. Is this the perfect moment you’ve been waiting to try LLMs? 💡</p><h4>HuggingChat by Hugging Face</h4><p>Hugging Face, an ML tools developer and AI code hub, launches its own open-source <a href="https://venturebeat.com/ai/hugging-face-launches-open-source-version-of-chatgpt-in-bid-to-battle-openai/">challenger to ChatGPT</a>. Hugging Face now serves only as a UI for the Open Assistant model but aims to grow into a chatbot client supporting many other models — so be sure to keep an eye on it! 🤗</p><h4>Building LLM apps for production</h4><p>What does it take to make your cool LLM app <em>production-ready</em>? ❓In her <a href="https://huyenchip.com/2023/04/11/llm-engineering.html">recent piece</a>, Chip Huyen talks about the ambiguity of natural languages, the stochastic nature of LLMs, and rapid evolution, then examines complex use cases involving tasks, agents, and control flows.</p><h3>News from the front lines of AI and MT</h3><ul><li><a href="https://slator.com/companies-launch-gpt-4-powered-machine-translation-services/">Companies Are Starting to Launch GPT-4-Powered Machine Translation Services</a></li><li><a href="https://www.miragenews.com/machine-translation-used-for-cuneiform-tablets-997834/">Machine Translation Used for Cuneiform Tablets</a></li><li><a href="https://news.tcnj.edu/2023/05/01/computer-science-asl-project/">Computer science research team explores how machine learning can translate sign language</a></li><li><a href="https://www.vice.com/en/article/4a3w3g/scientists-use-gpt-ai-to-passively-read-peoples-thoughts-in-breakthrough">Scientists Use GPT AI to Passively Read People’s Thoughts in Breakthrough</a></li><li><a href="https://www.theverge.com/2023/4/14/23683084/openai-gpt-5-rumors-training-sam-altman">OpenAI’s CEO confirms the company isn’t training GPT-5 and ‘won’t for some time’</a></li><li><a href="https://www.nytimes.com/interactive/2023/05/02/magazine/ai-gary-marcus.html">How Do We Ensure an A.I. Future That Allows for Human Thriving?</a></li><li><a href="https://www.cnet.com/videos/comparing-bing-chat-bard-chat-and-chatgpt/">Comparing Bing Chat, Bard Chat and ChatGPT (Video)</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f5a09900e3be" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/may-2023-mt-university-updates-ocr-in-the-mt-portal-new-languages-and-more-f5a09900e3be">May 2023: Machine Translation University updates, OCR in the MT Portal, new languages, and more</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[March 2023: Translate with GPT-4 and ChatGPT, New Plug-in for Trados Enterprise, and More]]></title>
            <link>https://medium.com/intento/march-2023-translate-with-gpt-4-and-chatgpt-new-plug-in-for-trados-enterprise-and-more-245e3f6bc575?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/245e3f6bc575</guid>
            <category><![CDATA[gpt-4]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[localization]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[James Hjerpe]]></dc:creator>
            <pubDate>Fri, 07 Apr 2023 07:00:00 GMT</pubDate>
            <atom:updated>2023-04-07T07:00:00.755Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dj9h--hsgqBDdGzBWIa7vA.jpeg" /></figure><p>With the start of a new quarter, we have some major updates to share with our community. Keep reading to learn more about translating with GPT-4 and ChatGPT through Intento, our new plug-in for Trados Enterprise, the best practices for training MT, and more. Enjoy!</p><p>🗣 Translate with GPT-4 and ChatGPT through Intento</p><p>🎉 The Intento plug-in for Trados Enterprise has arrived</p><p>🚀 7 new languages on the Intento platform</p><p>🎓 Best practices for training MT and mastering custom machine translation</p><p>💂‍♀️ Intento is coming to XTM Live in London</p><h3>Translate with GPT-4 and ChatGPT through Intento</h3><figure><a href="https://inten.to/api-platform/ai/text/translate"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ozNmpuH8DnU9zfr56PCSrg.png" /></a></figure><p>We have added ChatGPT and GPT-4 powered Machine Translation to the platform. We <a href="https://blog.inten.to/translating-with-gpt-4-the-latest-the-greatest-561025cb761c">tested the translation capabilities of GPT-4</a> and found them outstanding, while ChatGPT API (gpt-3.5-turbo) <a href="https://blog.inten.to/translating-with-gpt-4-the-latest-the-greatest-561025cb761c">performs on par with the top stock Machine Translation engines.</a></p><p>Try it now with our <a href="https://inten.to/api-platform/ai/text/translate">web demo</a> or access it through Intento plugins for Translation Management Systems. (Just keep in mind that GPT is a bit slow).</p><h3>Intento plug-in for Trados Enterprise</h3><figure><a href="https://appstore.rws.com/Plugin/199?utm_content=243325042&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024"><img alt="" src="https://cdn-images-1.medium.com/max/504/1*jfCm7pXJw-B-LtmZ1q4OvQ.png" /></a></figure><p>Intento MT Hub for Trados Enterprise integrates over 40 machine translation providers into a single plugin, letting you leverage the latest machine translation technology powered by <a href="https://www.linkedin.com/company/amazon/">Amazon</a>, <a href="https://www.linkedin.com/company/baidu-inc/">Baidu, Inc.</a>, <a href="https://www.linkedin.com/company/linkedin-com-company-deepl/">DeepL</a>, <a href="https://www.linkedin.com/company/google/">Google</a>, <a href="https://www.linkedin.com/company/microsoft/">Microsoft</a>, <a href="https://www.linkedin.com/company/naver/">NAVER Corp</a>, and <a href="https://www.linkedin.com/company/openai/">OpenAI</a>GPT. Key benefits include 10 machine translation providers supported via a single contract, 40+ machine translation providers, 750+ languages, source quality improvements, and automated post-editing.</p><p><a href="https://appstore.rws.com/Plugin/199?utm_content=243325042&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024">Try it out now</a> for a seamless and efficient machine translation workflow.</p><h3>7 new languages on Intento</h3><p><a href="https://www.linkedin.com/company/microsoft/">Microsoft</a> has added 13 new languages, one new to Intento, while <a href="https://www.linkedin.com/company/modernmt/">ModernMT</a> added 32 languages — 6 new to Intento.</p><p>Our tech is constantly scanning the machine translation landscape so that Intento users can access all available language updates. You can access all new languages on the <a href="https://inten.to/api-platform/ai/text/translate">Intento MT catalog</a>.</p><p>Here’s what’s new:</p><p><strong>Microsoft:</strong></p><p>ha — Hausa<br>ig — Igbo<br>ln — Lingala<br>nso — Northern Sotho/Sepedi/Pedi<br>ny — Chewa/Nyanja<br>rn — Kirundi/Rundi<br>rw — Kinyarwanda<br>sn — Shona<br>st — Sesotho/Southern Sotho<br>tn — Setswana/Tswana<br>xh — Xhosa<br>yo — Yoruba</p><p>lg — Ganda/Luganda (new to Intento)</p><p><strong>ModernMT:</strong></p><p>Acehnese — ace<br>Assamese — as<br>Bengali — bn<br>Bhojpuri — bho<br>Fijian — fj<br>Ilocano — ilo<br>Kamba — kam<br>Kikuyu — ki<br>Kongo — kg<br>Lingala — ln<br>Luba-Kasai — lua<br>Maithili — mai<br>Maori — mi<br>Marathi — mr<br>Mizo/Lushai — lus<br>Myanmar (Burmese) — my<br>Northern Sotho — nso<br>Chewa/Nyanja — ny<br>Odia (Oriya) — or<br>Pangasinan — pag<br>Punjabi — pa<br>Samoan — sm<br>Sindhi — sd<br>Sinhala (Sinhalese) — si<br>Sesotho/Southern Sotho — st<br>Yoruba — yo</p><p>Balinese — ban (new to Intento)<br>Banjar — bjn (new to Intento)<br>Buginese — bug (new to Intento)<br>Magahi — mag (new to Intento)<br>Minangkabau — min (new to Intento)<br>Plateau Malagasy — plt (new to Intento)</p><h3>New chapters on Intento Machine Translation University</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HUhsAbrqzv48XpGbA_r_Qg.png" /></figure><p>Two new chapters have been added to <a href="https://inten.to/machine-translation-university/">Intento Machine Translation University</a>.</p><p>Learn how to improve the quality of data used to train MT models to ensure the best performance. ‘<a href="https://inten.to/machine-translation-university/selecting-and-customizing-mt-engines-preparation-of-data-assets/">Best Practices for Training MT Models: Using Translation Memories and Glossaries</a>’ lets you know the data issues to address when working with translation memories.</p><p>Check out Chapter 5, ‘<a href="https://inten.to/machine-translation-university/mastering-custom-machine-translation-a-practical-guide/">Mastering Custom Machine Translation: A Practical Guide</a>’ to fully understand the training, evaluation, and optimization processes of custom MT systems.</p><h3>Intento at XTM Live in London</h3><figure><a href="https://xtmlive.com/"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nGgrE3z_h7NyQyFvsVDk2w.png" /></a></figure><p>Join Intento along with fellow localization peers and industry experts at <a href="https://xtmlive.com/">XTM Live 2023</a> in London from April 19th to April 20th.</p><p>On Day 1, Andy Nikulin (Director, Customer Success and Professional Services at Intento) and Wouter Maagdenberg (CEO &amp; Co-Founder at TXTOMedia) will be giving a talk around a use-case of MT with gender control that we’re working on with TXTO and CAREanimations. On Day 2, you’ll have the chance to get a live demo on MT Augmentation via XTM from Intento’s Product Director, Pavel Doronin.</p><p>Be sure to stop by booth #4 to hear about all the recent developments at Intento!</p><h3>News from the front lines of AI and MT</h3><ul><li><a href="https://slator.com/machine-translation-system-how-to-improve-chatgpts-translations/?utm_source=Slator+Newsletter&amp;utm_campaign=47eb3669be-Slator+Weekly+-+Dec+11%2C+2020_COPY_01&amp;utm_medium=email&amp;utm_term=0_28e0ad4ca5-47eb3669be-165537667&amp;mc_cid=47eb3669be&amp;mc_eid=21259a7966">How to Improve ChatGPT’s Translations</a></li><li><a href="https://multilingual.com/canva-create-translate/">Canva launches translation and other AI-powered features</a></li><li><a href="https://ai.googleblog.com/2023/03/universal-speech-model-usm-state-of-art.html?utm_content=240940084&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024">Google has announced their Universal Speech Model (USM)</a></li><li><a href="https://womeninlocalization.com/women-in-localization-calls-for-increasing-equity-on-international-womens-day/">Women in Localization Calls for Increasing Equity on International Women’s Day</a></li><li><a href="https://slator.com/speechless-recognition-can-ai-transcribe-language-its-never-heard/?utm_source=Slator+Newsletter&amp;utm_campaign=4383a5293e-Slator+Weekly+-+Dec+11%2C+2020_COPY_01&amp;utm_medium=email&amp;utm_term=0_28e0ad4ca5-4383a5293e-165537667&amp;mc_cid=4383a5293e&amp;mc_eid=21259a7966">Can AI Transcribe a Language It Has Never Heard?</a></li><li><a href="https://ut.ee/en/content/university-tartu-machine-translation-engine-now-supports-17-new-finno-ugric-languages?utm_content=242492280&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024">The University of Tartu machine translation engine now supports 17 new Finno-Ugric languages</a></li><li><a href="https://www.oecd-ilibrary.org/social-issues-migration-health/not-lost-in-translation_e1d1d170-en;jsessionid=4uBrIf5ou7ftN58kn075D7QHQtmrrk1khQapNjGH.ip-10-240-5-18">The implications of machine translation technologies for language professionals and for broader society</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=245e3f6bc575" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/march-2023-translate-with-gpt-4-and-chatgpt-new-plug-in-for-trados-enterprise-and-more-245e3f6bc575">March 2023: Translate with GPT-4 and ChatGPT, New Plug-in for Trados Enterprise, and More</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Translating with GPT-4: the Latest & the Greatest]]></title>
            <link>https://medium.com/intento/translating-with-gpt-4-the-latest-the-greatest-561025cb761c?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/561025cb761c</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gpt-3]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[gpt-4]]></category>
            <dc:creator><![CDATA[Konstantin Savenkov]]></dc:creator>
            <pubDate>Mon, 20 Mar 2023 20:51:29 GMT</pubDate>
            <atom:updated>2023-03-20T20:51:29.076Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1020/1*W5gtTqTvR9q9lAW5Q5oEqg.png" /></figure><p>Only a week after completing <a href="https://blog.inten.to/chatgpt-for-translation-surpassing-gpt-3-e8e34f37befb">our evaluation of ChatGPT (gpt-3.5-turbo) Machine Translation capabilities</a>, OpenAI kindly granted us beta API access to GPT-4. We promptly integrated it into our MT evaluation framework and conducted our initial tests.</p><h3>What’s new?</h3><p>Well, many things.</p><ol><li>Enhanced factual accuracy, although not crucial for translation, as the goal is to accurately translate the source sentence, even if it isn’t entirely factual.</li><li>Visual input compatibility, potentially enabling in-context translation, such as product descriptions using photos or UI elements using screenshots.</li><li>Larger prompt capacity, allowing up to 8K tokens (~6K words) in beta and 32K tokens (~24K words) in the full version. This could enable the full-text translation and submission of extensive information or metadata to guide translations, like glossaries, fuzzy TM matches, or comprehensive human-readable translation style guides.</li><li>Improved multilingual capabilities, which we’ll be exploring further in this article.</li></ol><p>Let’s dive in!</p><h3>Evaluation approach</h3><p>The general approach is the same one <a href="https://blog.inten.to/gpt-3-translation-capabilities-8a9290731a45">we have used to evaluate GPT-3</a>.</p><p>In <a href="https://blog.inten.to/gpt-3-translation-capabilities-8a9290731a45">our ChatGPT evaluation</a>, we used API and prompts supplemented with <a href="https://platform.openai.com/docs/guides/chat/introduction">system messages</a>. GPT-3.5 documentation suggests future models will pay greater attention to these engineering prompts, which might be true for GPT-4.</p><p>We tested both text generation (GPT-3-style) and chat completion (ChatGPT-style) prompts. As shown below, there aren’t many differences, making text generation a preferable option due to its shorter prompts.</p><h3>General domain translation</h3><p><strong>Surprise!</strong> GPT-4 achieves the highest average COMET score for English-to-Spanish translations. However, considering the confidence intervals (black ticks on the chart), this lead isn’t statistically significant, placing it in the same range as Google, GPT-3.5, Amazon, DeepL, Yandex, and Microsoft.</p><p>Determining the true leader would require human judgment, and we’ll share more details once we’ve had the opportunity to conduct such an evaluation.</p><figure><img alt="COMET scores for GPT-4 English to Spanish translation, general domain" src="https://cdn-images-1.medium.com/max/1024/1*YNb2F-ebwdykIPKlu9BGjQ.png" /><figcaption>COMET scores for GPT-4 English to Spanish translation, general domain</figcaption></figure><p>For General Domain texts in <strong>English to German</strong>, DeepL is still the best by far — however, GPT-4 has the highest scores in a group of the top-runners.</p><figure><img alt="COMET scores for GPT-4 English to German translation, general domain" src="https://cdn-images-1.medium.com/max/1024/1*hdpljqhmkf1sPrsYIif7xA.png" /><figcaption>COMET scores for GPT-4 English to German translation, general domain</figcaption></figure><h3>In-domain translation</h3><p>For the German in <strong>Legal</strong> and <strong>Healthcare, </strong>GPT-4 ends up being the best out of the GPT engines, however, still performs slightly worse than the first-tier engines:</p><figure><img alt="COMET scores for GPT-4 English to German translation, healthcare domain" src="https://cdn-images-1.medium.com/max/1024/1*ZPCF2t1Jm3mteP8XNCdgtQ.png" /><figcaption>COMET scores for GPT-4 English to German translation, healthcare domain</figcaption></figure><figure><img alt="COMET scores for GPT-4 English to German translation, legal domain" src="https://cdn-images-1.medium.com/max/1024/1*qBzxjR8OASpX6HzoTFs9tw.png" /><figcaption>COMET scores for GPT-4 English to German translation, legal domain</figcaption></figure><h3>Costs</h3><p>The 8K-token model available to us costs $0.03 per 1K tokens for the prompt and $0.06 per 1K tokens for completion. This equates to approximately $18 ($6 for the prompt and $12 for the completion) to translate 1M characters, making the cost similar to or slightly cheaper than Google Translate or DeepL.</p><p>However, it’s about 10 times more expensive than GPT-3.5-turbo, which costs $1.5 per million characters for machine translation.</p><h3>Technical performance</h3><p>Similar to <a href="https://blog.inten.to/gpt-3-translation-capabilities-8a9290731a45">what we’ve discovered for GPT-3</a>, GPT-4 experiences multiple API errors during translation. It also takes nearly twice as long as GPT-3 and GPT-3.5-turbo, averaging 46 minutes to an hour for a dataset of 500 segments. This results in a 5–7 second processing time per segment, almost 15–20 times longer than typical commercial engines.</p><p>The output may contain technical noise (like quotation marks), which is easily removable. In 0.5–1% of cases, the model returns an empty completion without a message, making retries essential.</p><h3>Conclusion</h3><p>GPT-4 boasts significant improvements in machine translation, including better factual accuracy, visual input compatibility, larger prompt capacity, and enhanced multilingual capabilities. Despite high average COMET scores in general domains, it falls short of specialized MT systems for in-domain translation. Its cost is similar to popular services like Google Translate or DeepL but 10x pricier than GPT-3.5-turbo.</p><p>While GPT-4’s technical performance lags behind commercial engines and occasionally encounters API errors, it shows great promise in machine translation. The rapid advancement in this well-known field highlights the ever-evolving technology landscape. Expect more sophisticated and efficient solutions from OpenAI and other industry leaders in the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=561025cb761c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/translating-with-gpt-4-the-latest-the-greatest-561025cb761c">Translating with GPT-4: the Latest &amp; the Greatest</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT for Translation: Surpassing GPT-3]]></title>
            <link>https://medium.com/intento/chatgpt-for-translation-surpassing-gpt-3-e8e34f37befb?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/e8e34f37befb</guid>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gpt-3]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Konstantin Savenkov]]></dc:creator>
            <pubDate>Tue, 14 Mar 2023 23:26:19 GMT</pubDate>
            <atom:updated>2023-03-18T07:40:52.336Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1020/1*MTHtx5dEfIslVvCgZ6NYOA.png" /></figure><p>We <a href="https://blog.inten.to/gpt-3-translation-capabilities-8a9290731a45">recently examined GPT-3’s (davinci-003 model) translation abilities</a> and found that traditional MT models still hold their place in AI. However, with the commercial release of ChatGPT (gpt-3.5-turbo) via API, now at only 10% of GPT-3’s cost, it’s time to reevaluate our conclusions.</p><h3>What’s new?</h3><p>People often mix up GPT-3 (commercial, API-accessible) with pre-API ChatGPT (free, no data confidentiality), similar to the confusion between Google Translate API and the free web-based Google Web Translator in 2016. However, GPT-3 and ChatGPT use distinct models.</p><p>The top GPT-3 model, davinci-003, launched in November 2022, while the ChatGPT model, gpt-3.5-turbo, debuted on March 1st, 2023.</p><p>Limited information exists on their differences, but gpt-3.5-turbo is about 10 times less expensive than GPT-3 (davinci-003) and more lenient with prompt engineering.</p><p>With the launch of commercial, API-enabled ChatGPT, <a href="https://openai.com/blog/introducing-chatgpt-and-whisper-apis">significant improvements in data protection and retention have emerged</a>, which are essential for the enterprise solutions we develop at Intento:</p><blockquote>Data submitted through the API is no longer used for service improvements (including model training) unless the organization opts in</blockquote><blockquote>Implementing a default 30-day data retention policy for API users, with options for stricter retention depending on user needs.</blockquote><blockquote>Removing our pre-launch review (unlocked by improving our automated monitoring)</blockquote><blockquote>Improving developer documentation</blockquote><blockquote>Simplifying our <a href="https://platform.openai.com/docs/usage-policies">Terms of Service and Usage Policies</a>, including terms around data ownership: users own the input and output of the models.</blockquote><p>General comparisons between the two models are <a href="https://scale.com/blog/chatgpt-vs-davinci">inconclusive</a>, as each has its strengths and weaknesses. However, gpt-3.5-turbo reportedly excels in zero-shot training (as a stock model), which is precisely what we need for basic Machine Translation!</p><h3>Wait, 10 times cheaper?</h3><p>Indeed, ChatGPT (gpt-3.5-turbo) costs $0.002 per 1K tokens, while GPT-3 is priced at $0.02 per 1K tokens.</p><p>But what about the cost of a million characters? Let’s see how <strong>GPT-4</strong> answers that (oops, a spoiler!).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/809/1*-fgq-uZ9u4-jI7tqN65dng.png" /></figure><p>Keep in mind, you pay for both the prompt and output, and prompts add characters to the input text. A realistic cost is roughly $1.5 per million characters, making it 13 times cheaper than standard Google Translate.</p><p>Can we expect high-quality translations at this price? <strong>Let’s find out!</strong></p><h3><strong>Evaluation approach</strong></h3><p>We used the same evaluation methodology as in our <a href="https://blog.inten.to/gpt-3-translation-capabilities-8a9290731a45">original GPT-3 Translation Capabilities post</a>.</p><p>There are various ChatGPT API prompt engineering approaches, such as <a href="https://platform.openai.com/docs/guides/chat/introduction">semi-structured prompts</a>.</p><p>We assessed multiple approaches and prompts for in-domain translation. While we didn’t find significant score differences among prompts, we did see varying numbers of risky and outlier translations. Thus, prompt engineering is crucial.</p><h3><strong>General domain translation</strong></h3><p>In the General domain, ChatGPT reached the top tier for <strong>English to Spanish</strong>, matching leaders like Google, Amazon, DeepL, Yandex, and Microsoft. This is a significant achievement, as GPT-3 only ranked in the second tier.</p><figure><img alt="COMET scores for ChatGPT English to Spanish translation, general domain" src="https://cdn-images-1.medium.com/max/1024/1*EW1PwqLAfqCscuwY33H0ig.png" /><figcaption>COMET scores for ChatGPT English to Spanish translation, general domain</figcaption></figure><p>For English to German, DeepL remains unmatched, but ChatGPT has advanced to the second tier, alongside the best alternative MT engines.</p><figure><img alt="COMET scores for ChatGPT English to German translation, general domain" src="https://cdn-images-1.medium.com/max/1024/1*j4d9iNnBb09-Ha1iEelNkg.png" /><figcaption>COMET scores for ChatGPT English to German translation, general domain</figcaption></figure><h3><strong>In-domain translation</strong></h3><p>In Legal and Healthcare, English to German, ChatGPT’s results have improved significantly but still trail top-ranking engines. This suggests that MT-specific model providers focus on maintaining more balanced and representative training data across various domains and content types.</p><figure><img alt="COMET scores for ChatGPT English to German translation, healthcare domain" src="https://cdn-images-1.medium.com/max/1024/1*5O0kh6F7Q_aF2U5-68jz8A.png" /><figcaption>COMET scores for ChatGPT English to German translation, healthcare domain</figcaption></figure><figure><img alt="COMET scores for ChatGPT English to German translation, legal domain" src="https://cdn-images-1.medium.com/max/1024/1*1UovGrP03jrZmzXOvsvj5Q.png" /><figcaption>COMET scores for ChatGPT English to German translation, legal domain</figcaption></figure><h3><strong>Business-critical errors and weak spots</strong></h3><p>Overall, ChatGPT’s output shares similar issues with other (good) MT engines. However, we notice that the number of business-critical translation errors varies significantly depending on the prompt (see below).</p><h4>English to Spanish, general domain</h4><figure><img alt="ChatGPT weak spots for English to Spanish translation (General domain)" src="https://cdn-images-1.medium.com/max/1024/1*I1E4JIr7965HsCIpDPv8-g.png" /><figcaption>ChatGPT weak spots for English to Spanish translation (General domain)</figcaption></figure><p>Common issues in the <strong>General</strong> domain:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/963/1*_yUuw3gqBX-1poLkCqdG4g.png" /></figure><h4>English to German, general domain</h4><figure><img alt="ChatGPT weak spots for English to German translation (General domain)" src="https://cdn-images-1.medium.com/max/1024/1*_1NbIIIPVLmtDmHvPbHjbw.png" /><figcaption>ChatGPT weak spots for English to German translation (General domain)</figcaption></figure><p>Common issues in the <strong>General</strong> domain:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/965/1*Z9Ov_IJoT9jxLChcKkzO9Q.png" /></figure><h4>English to German, Healthcare</h4><figure><img alt="ChatGPT weak spots for English to German translation (Healthcare domain)" src="https://cdn-images-1.medium.com/max/1024/1*ElzuygRzf27FMFASMtKY-A.png" /><figcaption>ChatGPT weak spots for English to German translation (Healthcare domain)</figcaption></figure><p>Common issues in the <strong>Healthcare</strong> domain:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/964/1*CmVZPZf6P8hV3FPiXNE3fQ.png" /></figure><h4>English to German, Legal</h4><figure><img alt="ChatGPT weak spots for English to German translation (Legal domain)" src="https://cdn-images-1.medium.com/max/1024/1*wo7NnmTzy136HdHzu4VZQw.png" /><figcaption>ChatGPT weak spots for English to German translation (Legal domain)</figcaption></figure><p>Common issues in the <strong>Legal</strong> domain:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*37vwoSocHzrGJnI9wFp_hA.png" /></figure><h3><strong>So, how to use it for translation?</strong></h3><p>Like any translation technology on the Intento platform, ChatGPT translation will be accessible through all Intento connectors, including 15 TMS systems and various Customer Service platforms like Salesforce, ServiceNow, and Zendesk.</p><p>Our ChatGPT connector is currently undergoing thorough quality assurance. To stay informed, <a href="https://inten.to/">register on the Intento website</a> and subscribe to our newsletter.</p><h3><strong>Conclusions</strong></h3><p>The technological progress in just under 3 months is astounding. <strong>ChatGPT performs on par with the best stock MT engines, at less than 10% of their cost.</strong></p><figure><img alt="COMET performance of GPT-3 and ChatGPT translation compared to commercial MT engines." src="https://cdn-images-1.medium.com/max/1024/1*2RxAaeYN_InPAStKYyQczw.png" /><figcaption>COMET performance of GPT-3 and ChatGPT translation compared to commercial MT engines.</figcaption></figure><p>Additionally, it offers morphological glossary application and few-shot learning capabilities, thanks to the versatile interface of this general model.</p><p>We’ll delve into the customizability of GPT-powered translation later, along with an evaluation of <a href="https://openai.com/product/gpt-4">GPT-4, which launched today, March 14th</a>.</p><p>Stay tuned for our upcoming State of the Machine Translation 2023 report later this year, where we’ll evaluate all GPT-based translation options. It’s going to be exciting!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e8e34f37befb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/chatgpt-for-translation-surpassing-gpt-3-e8e34f37befb">ChatGPT for Translation: Surpassing GPT-3</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[February 2023: Intento Machine Translation University, Using GPT-3 to Translate, and More]]></title>
            <link>https://medium.com/intento/february-2023-intento-machine-translation-university-using-gpt-3-to-translate-and-more-a7827803d454?source=rss----6d3731515c5e---4</link>
            <guid isPermaLink="false">https://medium.com/p/a7827803d454</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[machine-translation]]></category>
            <category><![CDATA[localization]]></category>
            <dc:creator><![CDATA[James Hjerpe]]></dc:creator>
            <pubDate>Thu, 02 Mar 2023 19:26:39 GMT</pubDate>
            <atom:updated>2023-03-02T19:26:54.360Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WQnwK4K-HowNA007XFod6Q.jpeg" /></figure><p>As we head into the last month of Q1, here’s a roundup of what turned into a very busy February for the Intento community. Keep reading to learn more about the launch of <a href="https://inten.to/machine-translation-university/">Intento Machine Translation University</a>, how to use <a href="https://inten.to/api-platform/ai/text/translate?utm_content=237695763&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">GPT-3 to translate with Intento</a>, a fresh plug-in to <a href="https://inten.to/machine-translation-university/selecting-and-customizing-mt-engines-infrastructure-and-workflows/?utm_content=239166077&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">enhance your employee experience via ServiceNow</a>, and more.</p><p>🎓 Intento Machine Translation University is here to help professionals go multilingual better, faster, and at scale</p><p>🤖 Use GPT-3 to translate with Intento</p><p>🤝 Intento Machine Translation Hub for ServiceNow for improved cross-team communication and employee satisfaction</p><p>🌎 Intento MT Hub for Trados Studios has significant updates in the RWS store</p><p>🗣 Systran adds English to Pashto</p><p>📚 Quick reads to keep you up to date</p><h3>Introducing Intento Machine Translation University</h3><figure><a href="https://inten.to/machine-translation-university/"><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HUhsAbrqzv48XpGbA_r_Qg.png" /></a></figure><p>We’re excited to announce the launch of<a href="https://inten.to/machine-translation-university/"> Intento Machine Translation University</a> — a program designed to help professionals in localization, customer service, and employee experience become proficient in modern technologies for automatic translation.</p><p>We’ll cover everything from NLP, NMT, and GPT to more advanced topics. With our program, you’ll learn how to assess the ROI and gain buy-in from stakeholders, explore the complexities of language technology, select, customize, and evaluate various options, and implement the technology to get the best return. With these quick lessons, you’ll be able to go multilingual faster, better, and at scale!</p><p>Click <a href="https://inten.to/machine-translation-university/">here</a> to get started.</p><h3>Use GPT-3 to translate with Intento</h3><figure><a href="https://inten.to/api-platform/ai/text/translate?u=237695763&amp;u=social&amp;u=facebook&amp;h=fbp-1037580486301217"><img alt="" src="https://cdn-images-1.medium.com/max/224/1*koPoXr9pRXX31P4r2uZSOg.png" /></a></figure><p><a href="https://www.linkedin.com/company/openai/">OpenAI</a>’s GPT-3 is finally available as a translation engine via Intento in <a href="https://www.linkedin.com/company/trados/">Trados</a>, <a href="https://www.linkedin.com/company/memoq-translation-technologies/">memoQ</a>, <a href="https://www.linkedin.com/company/xtm-international/">XTM International</a>, and other TMS services we work with, as well as in <a href="https://www.linkedin.com/company/salesforce/">Salesforce</a>, <a href="https://www.linkedin.com/company/servicenow/">ServiceNow</a>, <a href="https://www.linkedin.com/company/zendesk/">Zendesk</a>, and all other Intento integrations. We’ll even customize it on your data or have a custom prompt made for content generation. Try it right away with our free MT web demo <a href="https://inten.to/api-platform/ai/text/translate?u=237695763&amp;u=social&amp;u=facebook&amp;h=fbp-1037580486301217">here</a>.</p><h3><strong>Intento Machine Translation Hub for ServiceNow</strong></h3><figure><a href="https://store.servicenow.com/sn_appstore_store.do#!/store/application/a16cbf9d87ab011029392179cebb35d3"><img alt="" src="https://cdn-images-1.medium.com/max/588/1*uZPzIRE9meAQHx-Bz2MnOA.png" /></a></figure><p>Intento has announced the launch of the <a href="https://store.servicenow.com/sn_appstore_store.do#!/store/application/301e760a878719107c9221b8cebb3533/0.1.6">Machine Translation Hub for ServiceNow</a>, available for all ServiceNow users via its AppStore. The application helps enterprises scale global operations by building a seamless, language-inclusive workplace.</p><p>With improved source quality, automated post-editing (including tone of voice, consistent terminology, gender control, and other custom features), and direct access to 40+ machine translation engines, enterprises get the best possible machine translation, achieving 98% user acceptance.</p><p>Read more about enabling a multilingual experience for articles, chats, activity streams, and forms used in Employee Portal, Agent Workspace, and many other ServiceNow applications <a href="https://slator.com/intento-machine-translation-hub-for-servicenow-for-a-language-inclusive-workspace/">here</a>.</p><h3><strong>Intento MT Hub for Trados Studios has significant updates in the RWS store</strong></h3><figure><a href="https://appstore.rws.com/plugin/192/?utm_content=237827472&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024"><img alt="" src="https://cdn-images-1.medium.com/max/477/1*-YUI4zJbJeDDhBWlbsUc4w.png" /></a></figure><p>The update comes with Trados 2022 support, enhanced support of tags and HTML entities, and more minor improvements. We know that in-line tags and markups can be a real pain in most TMS systems, so we aim to normalize the markups to make them as understandable to the providers as possible. This way, post-editors do not lose focus on the language while fixing technical stuff in the translation editor.</p><p>Check out Intento MT Hub for Trados Studios <a href="https://appstore.rws.com/plugin/192/?utm_content=237827472&amp;utm_medium=social&amp;utm_source=linkedin&amp;hss_channel=lcp-10551024">here</a>.</p><h3><strong>Systran adds English to Pashto</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/399/1*kXpyxNImAnL9hFdGNJW04g.jpeg" /></figure><p>Systran has added English (en) to Pashto (ps), an Eastern Iranian language in the Indo-European language family. Our AI is constantly scanning the MT landscape for new changes, and you can see them all on our free MT web demo <a href="https://inten.to/api-platform/ai/text/translate">here</a>.</p><h3>News from the front lines of AI and MT</h3><ul><li><a href="https://ai.facebook.com/blog/large-language-model-llama-meta-ai/?utm_content=239325450&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">Introducing LLaMA: A foundational, 65-billion-parameter large language model</a></li><li><a href="https://openai.com/blog/introducing-chatgpt-and-whisper-apis">ChatGPT and Whisper APIs</a></li><li><a href="https://multilingual.com/what-other-creative-professions-can-learn-from-translation-about-the-effects-of-artificial-intelligence/?utm_content=238716725&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">What other creative professions can learn from translation about the effects of artificial intelligence</a>?</li><li><a href="https://illustroke.com/?utm_content=238894699&amp;utm_medium=social&amp;utm_source=facebook&amp;hss_channel=fbp-1037580486301217">New image generator: Illustroke</a></li><li><a href="https://multilingual.com/issues/february-2023/real-world-ai-driven-applications-are-signaling-a-new-frontier-in-mt/">Real-world AI-driven applications are signaling a new frontier in MT</a></li><li><a href="https://slator.com/large-language-models-get-better-machine-translation-of-rare-words/?utm_source=Slator+Newsletter&amp;utm_campaign=baa8e1a4a1-Slator+Weekly+-+Dec+11%2C+2020_COPY_01&amp;utm_medium=email&amp;utm_term=0_28e0ad4ca5-baa8e1a4a1-165537667&amp;mc_cid=baa8e1a4a1&amp;mc_eid=21259a7966">How Large Language Models Can Get Better at Machine Translation of Rare Words</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a7827803d454" width="1" height="1" alt=""><hr><p><a href="https://medium.com/intento/february-2023-intento-machine-translation-university-using-gpt-3-to-translate-and-more-a7827803d454">February 2023: Intento Machine Translation University, Using GPT-3 to Translate, and More</a> was originally published in <a href="https://medium.com/intento">Intento</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>