How AI is changing the future of translation management

Programs using machine learning can be faster, more accurate and less expensive when compared to the time and effort required by humans to do the same work. Machines and software systems that can “learn” and optimize efficiency are attractive tools for businesses everywhere, and the translation and localization industry is reaping the benefits of AI. Translation technology companies are now using AI-based rules engines to change the future of translation management by increasing the automation of workflows, vendor management and quality evaluation. There are a number of localization features that are possible with current and emerging AI technology.

Making translation workflows more agile

Translation technology can use AI in a few different ways to overcome the many challenges to translation workflows. Projects that take too long to get started, have too many manual processes and use multiple vendors all add complexity to the situation.

In theory, a workflow prediction engine can predict the appropriate workflow for newly-imported content. AI-enabled workflows can determine which combination of machine and human resources is best to optimize quality and accuracy. These intelligent workflows include phase automation, dynamic scheduling and active monitoring. Specifically, it is possible to do the following:

Phase automation: Phase automation determines which combination of machine and human resources is best to optimize quality and accuracy. It automatically selects and configures workflow phases based on document metadata (content type, campaign), due dates and other client-specific business intelligence.

Phase selection: After uploading a document, the dynamic workflow management engine automates the selection and configuration of each workflow step. Project task duration is calculated by subtracting the requested due date from the job submission date. The system then intelligently builds the appropriate workflow to meet the project due date without sacrificing quality. The system learns which workflow steps are mandatory versus which are nice to have. Client-specific business logic determines which phases will require more or less time to translate. In addition, metadata informs which phases can be excluded in order to meet the project deadline.

Phase configuration: Metadata not only drives phase selection, but also phase configuration, including automatic configuration of translation memory (TM) leverage (sequencing /prioritization or penalization), due dates, phase rules and assignments and more.

Dynamic scheduling: Dynamic scheduling calculates due dates for each phase of the workflow based on the time available to complete each step in the workflow. The algorithm gathers metadata such as due date, department, word count, content type and author, and then makes a prediction using its machine learning to automatically create the optimal workflow. It will take into consideration which language service providers (LSPs) are available, their content value, the translation levels — standard machine translation (MT), MT + post-editing, or something more curated — based on content type. It automatically assigns the appropriate teams or individuals based on content type and language. It also auto-calculates phase due dates and makes auto-assignments based on defined business logic.

Automatic assignment of staffing and resourcing: Phase automation includes using AI-driven processing so that staffing and resources are automatically assigned to vendors based on document metadata, target languages and job type. Vendor assignments can be individual or team-based, allowing any team-based assignments to be checked out by vendors, so that the project manager doesn’t have to manage vendor availability.

Active monitoring and alerts for projects that are at risk, late or past due: AI-based active monitoring recognizes how much content needs to be translated and can make on-the-fly adjustments to meet necessary deadlines. It reduces the need for routine, automated tasks, so project managers can perform tasks that are more valuable to the organization, like problem solving, responding to urgent issues or focusing on exception management.

Active monitoring identifies which steps can be added, skipped or canceled to meet the desired deadlines. As a part of dynamic workflow management, the software can auto-cancel a nonmandatory phase that is past due.

If a workflow phase is in danger of missing its due date, its status could be set to “at risk” so project managers and assignees can see the status and take action. If an incomplete workflow phase misses its due date, its status is set to “past due.” Upon completion, a phase with a status of “past due” will be set to a “late” status.

AI is bringing more flexibility and scalability to translation workflows by reducing the number of manual processes and reducing project turnaround times. It also allows project managers to increase translation speed without sacrificing quality.

Automating vendor


Language services vendor management is a complex management task. It requires vetting multiple LSPs, requesting multiple bids and comparing different rate structures. It can include literally hundreds of projects that all require monitoring and managing to ensure on-time delivery. Adding to the complexity, LSPs typically use several different computer-assisted translation (CAT) tools and maintain multiple linguistic assets in various offline locations.

Translation software can bring AI-driven automation and multilingual business intelligence to translation management. The entire process for managing vendors — vendor selection, tracking costs and spending, vendor performance — can be easier and more automated with AI algorithms. With enough data, organizations can easily and repeatedly select vendors that provide the highest translation quality and that consistently deliver jobs on time. The features that are possible with this technology are these:

Integrated and automated vendor selection: AI-driven vendor management simplifies and consolidates the process for requesting quotes, setting rates and pricing, choosing vendors, managing deadlines, tracking spending and measuring translator quality and performance. This can give project managers insight into how to better manage workloads and resources for maximum throughput.

Centralized tracking of rates: Smart software can also automate many of the steps required for creating a purchase order and closely tracks translation spending. It can also track the leveraging of TMs to gauge the efficient reuse of linguistic assets across the enterprise.

Automatic cost calculation: AI-based vendor management includes auto-calculation of costs even when specific jobs have been skipped or canceled. A project manager can manually skip or cancel a phase, target or entire document. With active monitoring, jobs can also be auto-skipped or auto-canceled in order to ensure on-time delivery. When this happens, the AI-driven vendor management system is able to proactively alert vendors of the skipped or canceled job, ensure that additional work cannot be performed on those skipped or canceled jobs, and then automatically calculate the the costs for the work that was completed before the job was canceled. This makes invoicing easier, as project managers and vendor managers no longer have to worry about notifying vendors of changes made to the project midstream, or figuring out how much work was done after the fact in order to manually calculate their costs.

Monitoring translation quality

Monitoring translation quality requires time, effort and a great degree of localization expertise. Even the most globalization-savvy enterprises with internal localization departments and dedicated quality managers still may resort to spreadsheets and labor-intensive manual processes to evaluate linguistic quality. There haven’t been many good tools on the market for creating quality programs at scale. However, new quality management apps can use quality data and machine learning throughout the entire translation process in order to bring machine learning to questions of quality:

Real-time application of quality standards: A linguistic quality evaluation app can evaluate quality using a model of predefined values, parameters and scoring based on representative sampling. It gives project managers or quality managers the ability to automate application of their preferred quality program — an industry standard such as DQF or a customized quality program — to meet their exact needs. An AI-driven quality app creates a foundation for trigger-based automation, rule-driven systems and data collection.

Flexible scoring by content type: Not all content requires the same quality level. AI-based scoring knows that a discussion forum or blog post will not need the level of review that a legal document or customer-facing brochure might require. Flexible and dynamic scoring enabled by an algorithm can set up scorecards automatically depending on content type.

Identification of error types and severity levels: Error codes and scoring algorithms determine how to score and identify error types. Thus, project managers gain insight into translation accuracy, spelling, instances of overtransalation and undertranslation. It differentiates severity of the translation error by adding a numerical multiplier, ranking severity levels from 1-4 or by any method they choose.

Standardized word count algorithm: An algorithm for establishing a standardized word count creates a baseline for comparing quality scores among documents of different sizes for an apples-to-apples comparison. This displays quality scores and grades across documents of varying sizes.

Programmatic language quality assurance: Some translation management system workbenches include programmatic quality checks, systematic quality checks that catch errors such as missing format tags, missing terms, punctuation differences, malformed or missing IP addresses, as well as email addresses. It can detect hundreds of error types, allowing linguists to quickly navigate to the error, review and correct potential quality issues before delivering the completed translation.

Analysis of translation quality over time: Quality apps can also collect data so managers can filter by quality scores for all documents in a project to see only those failing to meet the assigned quality standard. It can be used to analyze trends, view quality scores over time by project and by locale, for all targets. This allows identification of the most common mistakes. Pie charts can display the number of quality issues in each category such as terminology, style, language and accuracy. It gathers objective data for insights into improving quality delivery.

LSPs can now handle workflows, vendor management and quality evaluation all with AI-based technology. As a result, managers can focus on customer relationships and invest more time in business development strategies. These types of activities leave the mundane to the machine and those requiring a human touch to the vendor managers.

Using AI will trigger a revolution in the way the language services industry manages translation. But the benefits gained by automating, collecting data and assigning tasks based on specific criteria is worth entering this brave new world of AI-based translation management.