{"id":92272,"date":"2024-10-02T10:24:18","date_gmt":"2024-10-02T08:24:18","guid":{"rendered":"https:\/\/phrase.com\/?p=92272"},"modified":"2024-11-13T17:14:50","modified_gmt":"2024-11-13T16:14:50","slug":"98-where-machine-beats-memory-in-translation-speed","status":"publish","type":"post","link":"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/","title":{"rendered":"98: Where Machine Beats Memory in Translation Speed"},"content":{"rendered":"\n<div id=\"acf\/text-block_934a3272069950d3a42ceff7da41d0f0\" class=\"pxblock pxblock--text alignfull spacing--default bg--white\">\n\n\t\n\t<div class=\"container\">\n\t\t<div class=\"wysiwyg animate-in\">\n\t\t\t<p><span style=\"font-weight: 400;\">In our hyper-connected world, every second matters. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">With AI driving the explosion of content, the speed at which information is delivered is crucial for staying competitive. Communicating in a language your audience understands is vital for reaching global markets\u2014but how can we accelerate this process? <\/span><\/p>\n<p><span style=\"font-weight: 400;\">We decided to find out.<\/span><\/p>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_69_1 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Overview<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#crunching-the-numbers\" title=\"Crunching the numbers\">Crunching the numbers<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#what-exactly-do-we-mean-by-%e2%80%9cediting%e2%80%9d-time\" title=\"What exactly do we mean by \u201cediting\u201d time?\">What exactly do we mean by \u201cediting\u201d time?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#what-about-the-coffee-breaks\" title=\"What about the coffee breaks?\">What about the coffee breaks?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#why-not-just-use-edit-distance\" title=\"Why not just use edit distance?\">Why not just use edit distance?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#which-segments-made-the-cut\" title=\"Which segments made the cut?\">Which segments made the cut?<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#the-curious-case-of-the-flat-mt-line\" title=\"The curious case of the flat MT line\u00a0\">The curious case of the flat MT line\u00a0<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#so-what-does-this-mean-for-you\" title=\"So, what does this mean for you?\u00a0\">So, what does this mean for you?\u00a0<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-3'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/phrase.com\/blog\/posts\/98-where-machine-beats-memory-in-translation-speed\/#analytics-your-superpower-for-localization-insights\" title=\"Analytics, your Superpower for Localization Insights\">Analytics, your Superpower for Localization Insights<\/a><\/li><\/ul><\/nav><\/div>\n<h3><span class=\"ez-toc-section\" id=\"crunching-the-numbers\"><\/span><b>Crunching the numbers<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">We crunched a ton of data to figure out how customers can cut down on the time they spend translating content. <\/span><b>956,675,510 data points<\/b><span style=\"font-weight: 400;\"> to be exact. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">As you&#8217;d expect, segments with higher fuzzy match scores are quicker to edit. But once you hit about 93, things level out\u2014below that, editing times don&#8217;t really get faster, since those TM segments tend to be pretty much rewritten from scratch.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Then, we compared editing times for Translation Memory (TM) segments and Machine Translation (MT) segments. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">On average, editing MT-generated segments takes 5.8 seconds.<strong> The only TM segments that beat that are the ones with a score of 98 or higher<\/strong>. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">So, unless you&#8217;re dealing with super-high fuzzy matches, machine translation is your best bet for saving time.<br \/>\n<\/span><\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-92308\" src=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-B-300x150.png\" alt=\"Graph depicting editing time of translation segments in relation to fuzzy match scores. TM editing times vary across different scores, while MT is represented by a flat line, indicating a consistent editing time regardless of match score.\" width=\"862\" height=\"431\" srcset=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-B-300x150.png 300w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-B-768x384.png 768w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-B.png 1025w\" sizes=\"(max-width: 862px) 100vw, 862px\" \/><\/p>\n<h3><span class=\"ez-toc-section\" id=\"what-exactly-do-we-mean-by-%e2%80%9cediting%e2%80%9d-time\"><\/span><b>What exactly do we mean by \u201cediting\u201d time?<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">We actually track two times, <\/span><b>thinking time<\/b><span style=\"font-weight: 400;\"> and <\/span><b>editing time<\/b><span style=\"font-weight: 400;\">. When a user clicks into a segment, both thinking and editing time begin to be tracked. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Once the user starts editing the segment, the counting of thinking time stops, but editing time continues to be recorded. When the user then clicks into a different segment, the counting of editing time stops.<\/span><\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-92302\" src=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-A-300x150.png\" alt=\"Illustration showing the flow of thinking time and editing time in translation. It visualizes time spent clicking into a segment, starting typing, and clicking into a different segment, with distinct sections for thinking time and editing time.\" width=\"910\" height=\"455\" srcset=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-A-300x150.png 300w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-A-768x384.png 768w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1024-image-A.png 1025w\" sizes=\"(max-width: 910px) 100vw, 910px\" \/><\/p>\n<p><span style=\"font-weight: 400;\">When a user re-enters a previous segment, the counting resumes and any new times are added to the previous values. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">If there is any editing upon re-entering, both thinking and editing times are updated. However, if re-entering a segment results in no editing, the times remain unchanged despite the segment being re-confirmed.<\/span><\/p>\n<h3><span class=\"ez-toc-section\" id=\"what-about-the-coffee-breaks\"><\/span><b>What about the coffee breaks?<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">There\u2019s always that set of segments with unusually long editing times. These often happen when the linguist gets distracted mid-edit\u2014maybe they\u2019re checking their phone or getting a coffee. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Since the average editing time is around 20 seconds, we add a 50% buffer and remove any segments that take longer than 30 seconds. Sure, some longer edits might be legit, but most of the time, it\u2019s just noise.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">We tested whether cutting out these 30-second-plus segments affects the data. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">On larger datasets, it barely makes a difference. But on smaller sets, it can be more noticeable. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">This makes sense\u2014if you take a &#8220;coffee break&#8221; in the middle of a small job, it\u2019s going to have a pretty obvious impact on the total editing time. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">That\u2019s why we filter out those long segments to keep the data clean.<\/span><\/p>\n<h3><span class=\"ez-toc-section\" id=\"why-not-just-use-edit-distance\"><\/span><b>Why not just use edit distance?<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">The edit distance metric can be pretty misleading\u2014it makes fixing a translation seem way easier than it actually is. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">It simplifies everything down to just counting keystrokes for corrections, but completely ignores the brainpower and expertise required to make sure the translation is actually accurate. <\/span><\/p>\n<p><b>What edit distance really misses is the thinking time.<\/b><span style=\"font-weight: 400;\"> It\u2019s not just about hitting the right keys. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Editing time, while not perfect, does a much better job of capturing that mental effort and complexity. It reflects the real work involved, not just the number of keystrokes.<\/span><\/p>\n<h3><span class=\"ez-toc-section\" id=\"which-segments-made-the-cut\"><\/span><b>Which segments made the cut?<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">To ensure an accurate view of job performance, we include all segments from completed workflows, even those with no editing time. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">This helps to reflect the true impact of optimization. These are the segments we consider:<\/span><\/p>\n<ul>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">We\u2019re pulling in segments from jobs where all workflow steps are completed. We treat job completion as a sign that the segments are &#8220;done&#8221;.<\/span><\/li>\n<li style=\"font-weight: 400;\" aria-level=\"1\"><span style=\"font-weight: 400;\">We include segments with 0 editing time, because if optimizing leads to more segments that don\u2019t need any edits, that needs to be factored in.<\/span><\/li>\n<\/ul>\n<h4><strong>Example:<\/strong><\/h4>\n<p><span style=\"font-weight: 400;\">Let\u2019s say you have a 5-segment job, and originally 3 segments needed edits with times of 6, 8, and 10 seconds. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">After optimizing the TM Threshold, now only two edits are needed\u20148 and 10 seconds. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">If we only looked at edited segments, the average time would have gone from 8 seconds before to 9 seconds after, which wrongly suggests things got worse. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">But when we include the untouched segments, the averages are 4.8 seconds before and 3.6 seconds after, which reflects the improvement.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">For the same reason, we don\u2019t exclude locked segments either.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Data for segments from shared jobs will only be visible to the organization that created the job (the buyer).<\/span><\/p>\n<h3><span class=\"ez-toc-section\" id=\"the-curious-case-of-the-flat-mt-line\"><\/span><b>The curious case of the flat MT line\u00a0<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">We\u2019re correlating TM editing times with fuzzy scores, but for MT suggestions, fuzzy scores don\u2019t exist\u2014they\u2019re not applicable. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Instead, we have QPS scores (Quality Performance Scores) for MT. Figuring out if there\u2019s a link between fuzzy scores and QPS scores (so we can put them both on the same X-axis) is something we\u2019ll explore in future research. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">We\u2019ll update you on that later. For now, we\u2019re just using the overall average editing time for MT suggestions, which is why the line appears flat.<\/span><\/p>\n<h3><span class=\"ez-toc-section\" id=\"so-what-does-this-mean-for-you\"><\/span><b>So, what does this mean for you?\u00a0<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">The chart above shows averages across all data from all organizations. On October 2nd, we added a personalized dashboard to the <a href=\"https:\/\/phrase.com\/platform\/analytics\/\">TMS Phrase Analytics<\/a> with data specific to your organization. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">You are able to spot where the orange and purple lines intersect. <strong>This crossover point marks the golden TM threshold<\/strong>, helping you optimize for maximum effort savings.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Of course, different content types need different approaches. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">Translating marketing content is a whole different ball game compared to technical documentation. The dashboard will also let you filter the data using predefined options to fine-tune the results to your needs.<\/span><\/p>\n<p>&nbsp;<\/p>\n<h3><span class=\"ez-toc-section\" id=\"analytics-your-superpower-for-localization-insights\"><\/span><b>Analytics, your Superpower for Localization Insights<\/b><span class=\"ez-toc-section-end\"><\/span><\/h3>\n<p><span style=\"font-weight: 400;\">Analytics is like the business world&#8217;s superhero, always swooping in to save the day by spotting where time and money are leaking. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">We\u2019re excited to take our Phrase Analytics to the next level with these new, actionable insights. <\/span><\/p>\n<p><span style=\"font-weight: 400;\">This is our first step into using ultra-granular, segment-level metadata. While this initial version might not be flawless, we&#8217;re committed to working closely with our customers to refine it and deliver tools that give them real oversight over their localization processes.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">Stay tuned.<\/span><\/p>\n<p><span style=\"font-weight: 400;\">\u00a0<\/span><\/p>\n\t\t<\/div>\n\t<\/div>\n<\/div>\n\n\n\n<div id=\"acf\/blog-cta-block_9dfacb71f1c40b08f6d66c6523a96972\" class=\"pxblock pxblock--blog-cta alignfull bg--outline image--orientation-landscape\">\n\t<div class=\"block-container\">\n\t\t\t\t\t<div class=\"image image--align-top\">\n\t\t\t\t<img loading=\"lazy\" decoding=\"async\" width=\"2601\" height=\"1329\" src=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time.png\" class=\"attachment-original size-original\" alt=\"\" srcset=\"https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time.png 2601w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time-300x153.png 300w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time-1024x523.png 1024w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time-768x392.png 768w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time-1536x785.png 1536w, https:\/\/phrase.com\/wp-content\/uploads\/2024\/09\/P1019-TM-Threshold-Webinar-Carousel-image-1300-X-666-no-date-or-time-2048x1046.png 2048w\" sizes=\"(max-width: 2601px) 100vw, 2601px\" \/>\t\t\t<\/div>\n\t\t\t\t<div class=\"content\">\n\t\t\t<p><span style=\"color: #1080fc;\">ON-DEMAND WEBINAR<\/span><\/p>\n<div class=\"c-message_attachment__row\"><span style=\"color: #181818;\"><strong><span class=\"c-message_attachment__title\" data-qa=\"message_attachment_title\"><a class=\"c-link c-message_attachment__title_link\" style=\"color: #181818;\" href=\"https:\/\/phrase.com\/resources\/webinars\/breaking-the-70-barrier-tm-threshold-optimization\/\" target=\"_blank\" rel=\"noopener noreferrer\" data-qa=\"message_attachment_title_link\"><span dir=\"auto\">Breaking the 70% Barrier: A New Approach to TM Threshold Optimization<\/span><\/a><\/span><\/strong><\/span><\/div>\n<p class=\"c-message_attachment__row\"><span class=\"c-message_attachment__text\" data-qa=\"message_attachment_text\"><span dir=\"auto\">Tune in to our game-changing webinar where we challenge the industry\u2019s reliance on the 70% Translation Memory (TM) threshold.<\/span><\/span><\/p>\n<p><strong><a href=\"https:\/\/phrase.com\/resources\/webinars\/breaking-the-70-barrier-tm-threshold-optimization\/\">Watch now &gt;&gt;&gt;<\/a><\/strong><\/p>\n\t\t<\/div>\n\t<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>Discover how machine translation outperforms traditional translation memory in speed and efficiency. Learn how to optimize your localization process with data-driven insights and analytics.<\/p>\n","protected":false},"author":72,"featured_media":92317,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_stopmodifiedupdate":false,"_modified_date":"","_searchwp_excluded":"","footnotes":""},"categories":[41,42],"class_list":["post-92272","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-translation","category-phrase-and-beyond"],"acf":[],"_links":{"self":[{"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/posts\/92272"}],"collection":[{"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/users\/72"}],"replies":[{"embeddable":true,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/comments?post=92272"}],"version-history":[{"count":15,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/posts\/92272\/revisions"}],"predecessor-version":[{"id":94793,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/posts\/92272\/revisions\/94793"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/media\/92317"}],"wp:attachment":[{"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/media?parent=92272"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/phrase.com\/wp-json\/wp\/v2\/categories?post=92272"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}