MultiTrans Prism

In May 2011, MultiCorpora announced its next-generation product, MultiTrans Prism. The company showcased Prism at the Localization World Conference in June, and at the time of this writing in August, MultiCorpora has begun a controlled release of the Prism product line to key customers. By the time this article appears in October, Prism should be available for general release.

MultiTrans Prism is positioned as a fully-featured translation management system (TMS). Combining features from the latest release of MultiTrans, upgraded server technology, and the translation project management technology acquired from Beetext, MultiTrans Prism now brings together the three vital components of TMS technology: project management (PM), translation memory (TM) and terminology management. While the guiding concept of Prism is that the whole package is greater than the sum of its parts, MultiCorpora stresses that the parts can also function independently. The PM component that is the subject of this review is the best example of independent functionality; MultiCorpora continues to sell it on a standalone basis in the market segment formerly addressed by Beetext.

Because just six months ago I reviewed the MultiTrans TextBase TM component that was folded into the Prism release (MultiLingual #119 April/May 2011), I will concentrate the focus of this article on how the product components integrate with one another, following a project from initial request to delivery.

To set the stage, let’s envision a business with an in-house project manager and several in-house translators that outsources 60% of its translation projects. For this discussion, we will restrict the scope to just several steps. We will not include process steps such as revision and quality assurance.

Translation projects usually begin with a request for services from a customer via a web portal that is called an Online Request Form in Prism. The form captures all the necessary data required for the project, including who the requestor is, what languages are required, the actual source language documents, reference documents, plus invoicing, delivery data and any special instructions. Online Request Forms may be configured and customized to suit particular customers by a Prism system administrator. For example, if a customer frequently requires a particular language combination or a certain project manager, these may be configured as the default. Specific translation workflows may be defined as well and associated with the customer, the language combination or other recurring project attributes. As soon as the customer defines the attributes and sends the translation request into the system, the predetermined unique workflow can automatically kick in to transport the project through the process.

We will now switch our view to that of the project manager (Figure 1). A project manager automatically receives a notification that a request has been initiated by the customer. At this point, if a workflow that corresponds to the attributes of the project is already in place and associated with the customer, some tasks may have already been triggered automatically. Some of these are Flow Machine Tasks. Currently, there are three such tasks available for workflows: Analysis, Pretranslation and Export from XLIFF. This means that if a translation request is issued outside of working hours or while the project manager is otherwise occupied, some of the project preparation work will already be finished. Prism offers a choice of two background file format technologies for pre-processing files: either Microsoft Word or XLIFF, which can be used directly in the Prism XLIFF Editor or in another computer assisted translation tool of the translator’s choice as long as the tool adheres to the XLIFF standard.

If no workflow has been associated with the customer in advance, the project manager may now either apply a pre-existing one or create a customer workflow from the request. A typical workflow might include steps such as analysis, pretranslation, translation, editing, proofing, regeneration of the original document format from the intermediate XLIFF, validation, final delivery and TextBase updating. Of these, translation, editing, proofing, validation and final delivery require human interaction. Analysis, pretranslation, regeneration and TextBase updating can be automated if MultiTrans Prism is operating in the background as the server solution.

Pretranslation in Prism transfers full and fuzzy matches as well as terminology from the TextBase TMs and TermBases, and when a subsegment match is found in the source text, it will be highlighted (Figure 2). This allows translators to leverage Multi-Trans’ Fetch from Web functionality to retrieve subsegment translation suggestions from the server-based TMs if they so desire.

An additional option for pretranslation is to consult publicly-available memory stores or machine translation to retrieve segment matches that are not found in its own 
TextBase TM. In this way, it is possible to output a completely pretranslated document that either complies with the “good enough” standard or can be post-edited, depending on the customer’s wishes.

As a workflow is built, tasks and providers are added. Document statistics are automatically propagated from the original analysis into the individual tasks, as appropriate. If the project consists of numerous target languages for which the workflows differ somewhat, the project manager can copy an existing workflow and then modify it to suit the particular language circumstances. When the workflow is fully built, the project manager saves it. This triggers the first task. Prism sends a notification to the provider that a task is available to start and updates the lists of pending tasks of other service providers in the workflow. This way, all providers in the workflow are forewarned that they have tasks queued in their work pipeline.

We will now once again switch views, this time to that of the service provider to whom the task is assigned. Upon receiving the notification, the providers can log onto Prism and download the working files for their task. If they receive an XLIFF file, they can use an editing environment of their choice. If connection to the Prism TextBase TM and TermBase servers is desired, either the XLIFF Editor or Microsoft Office products may be used together with MultiTrans plus a small built-in component called the Translation Agent. Translators who use their own editing environments or the XLIFF Editor in isolation may do so without the benefit of MultiTrans and the Translation Agent; if they are working on a pretranslated document, they can still flesh out their translations with content ranging from subsegments to full paragraphs pulled from the centralized TextBase server with functionality called Fetch from Web.

Tasks are completed by uploading the finished documents back to Prism through the provider portal and marking the task as complete. This triggers a notification for the next service provider in line, who may now download those documents. And so it goes, from provider to provider, until all tasks have been completed. The project manager can view the current status of all tasks in the workflow at any time. After the production steps are completed and the project is ready for delivery, the service requestor receives a notification that documents are ready to be downloaded. They log on and retrieve the finished work by downloading it. Other means of transmission — e-mail, for example — may also be used if required.

After work has been delivered, invoices may be created directly within Prism. Invoice composition makes use of the data that is already available in the system, so it is quick, easy and accurate. Prism supports the accumulation of data from multiple tasks on one invoice. In this way, creditors could assemble data from all tasks delivered during one month instead of billing task-by-task. However, a workflow program that captures financial data should not be confused with a fully-featured accounting solution, even if comprehensive statistical reporting is available. To this end, Prism can exchange data with third-party accounting packages that are specifically designed for that. For example, Prism can both push data directly to or retrieve data directly from QuickBooks without using an intermediate file format. Interaction with other accounting software programs may be achieved via CSV export.

One might ask what happens if a provider creates an invoice that deviates from the project data. There are several answers to that question. One answer is that Prism’s invoicing module allows project managers to reject invoices, sending them back to providers for rework and subsequent resubmission. Another answer lies in Prism’s contract feature. A contract can be created from project data prior to commencement of the workflow. In this case, Prism actually prevents the provider from creating invoices that exceed the values specified in the contract.

Even after invoicing, a project is not complete without updating the TMs. This can also be automated. Because updating of the memories occupies the TextBase TM server, Prism supports scheduling the update action at a specific time when the server is less likely to be loaded with transactions. So, when a job is completed and TextBase TM updating is desired, what actually happens is that the update job goes into a queue to wait for the scheduled update time. If the update should be carried out right away, for example, to make the new TM content available to other translators in an ongoing project, this may also be done. In other situations, shared TM models such as concurrently available team memory might be more appropriate.

We should take a quick peek behind the scenes at a couple of typical usage scenarios to see how workflows are designed and implemented. Users with appropriate rights can access the Workflow Templates page (Figure 3). Here, they will find a graphical representation of the workflow on the top half of the page and a step detail panel on the lower half. Steps may be added or deleted, workflow paths split and rejoined, and tasks with refusal rights built in. Tasks can even be reiterated in a looping manner that closely mirrors the reality of a typical localization project.

Another issue that commonly comes up in practice is the handling of updates from customers. This is frequently a big annoyance for translation service providers: a document has entered the translation workflow, and then the customer sends in a slightly changed version with the same file name. Prism handles it as a new version of a document that already exists in the workflow that can be flagged specifically as such. Internally, it will be kept separate from the previous version, and the assigned providers will receive notification that the document has changed. There will also be a special marking next to the project name in the project manager’s list of requests-in-progress that indicates that a document contained in the request has been modified, and a version history will be generated. This version history may be viewed by the project manager in a window, and the most recent version will now be available for download by the provider. There is flexibility regarding who receives notifications about the update; for example, it can be just the project manager, or the project manager and all providers who have started their tasks. Also, various triggers can be set that initiate update notifications.

Another feature worth mentioning is Pooling. This feature allows for the creation of a pool of translators who can be assigned to a customer or a request. If this is the case, then all of the translators in the pool will receive notification that work is available. The first translator to respond will then be assigned the task, and it will disappear from the work queues of the other translators in the pool. Or, optionally, all translators in the pool can be allowed to express interest in taking the assignment, and the project manager can choose which one to give it to.

The final feature that I would like to discuss is the Find module. This is essentially a mechanism optimized for searching through unstructured text. Translators sometimes find themselves tasked with work and know that they have previously translated something similar. But they don’t remember what it was or for whom it was done. Find provides the ability to search through the repository of all source, final target and reference documents based on keywords in order to locate documents that correspond to the entered search criteria. While this sounds similar to TextBase TM insofar as it relies on indexing of document content, it serves a different purpose. TextBase TM is capable of returning matching subsegments, segments, paragraphs, or even whole documents during pretranslation or translation. Find points users to actual documents in the project directory (as opposed to content in the TextBase) that have been created in the past and, more specifically, to useful metadata such as the translation order number. I can imagine how convenient this search function would be, especially when translators work on many short documents year in and year out.

MultiCorpora previously offered integration with workflow that functioned as a separate application. By offering workflow and PM components unified together with linguistic services within Prism, MultiCorpora has now provided project managers with an easier way of handling large, multilingual translation projects that does not require switching tools or seeking support from two tools vendors.

I liked the ability to use automation to create XLIFF files for use in other translation tools. In experimenting with pretranslated output, however, I found that it is necessary to jump through a few hoops to achieve interoperability. By default, Prism uses ISO 639-3 (three-character) language codes. In their respective default configurations, neither SDL Trados 2009 nor MemoQ likes these codes (SDL Trados 2007 was OK). MemoQ is more compliant than SDL Trados 2009; it offers a setting that allows ignoring the language codes when importing. After choosing this setting, import into MemoQ was successful. A workaround is to set up the language codes in Prism to use the 2-character codes, although this involves some initial manual labor. On the return trip there is no problem due to the fact that Prism’s internal format is native XLIFF.

The XLIFF standard is flexible regarding language code designation, so one cannot point a strict finger at any of the aforementioned tool providers; this flexibility in the standard does come at a cost of out-of-the-box user friendliness. On the other hand, Trados additionally blocked import based on what it determined to be an illegal XLIFF tag. My research revealed that the tag is, in fact, legal, so I feel that some finger-pointing is justified here.

Import of Prism’s pretranslated content into SDL Trados 2007 can be successfully accomplished when using the MS-Word pretranslation option. In this case, Prism offers an option to create a bilingual SDL Trados Word file. I appreciated the way in which workflows can be designed as well as the way workflows can be set up with loops and refusals. This is a usage scenario we all have seen in real translation and localization projects.

However, talking about workflows and automation, I feel, as with automation features elsewhere, that it cannot be stressed enough that having a workflow does not solve every issue. A project manager needs to know exactly what the workflow does — for example, where the matches will come from. Although it is useful that default matching settings can be configured to be analogous to those used in MultiTrans, these same settings cannot be overridden at the project level. At the project level, it is currently only possible to determine the TextBase and TermBase used for analysis and pretranslation, as well as the TextBase to be updated.

Finally, I noticed that Prism allows for the entry of improper data in some fields without real-time corrections. This is inconvenient, as users must go back to make the corrections before continuing. As with all software that I have reviewed so far, I would like to see MultiCorpora implement more error messages to help users rectify their mistakes directly during the data entry process. This includes a little more guidance for users, especially project managers new to the job or the tools, to help them along in using the workflow component — maybe some more popup texts with explanations covering what needs to go into what field or what other settings need to be made if one option is ticked and so on. But all in all, Prism is a development in the right direction; it is a tightly integrated tool set that also allows separate use of the individual components.