MemoQ is arguably one of the best CAT tools you can buy today. Since its introduction in 2006, it has constantly posed a threat to SDL’s Trados lineup virtual monopoly over the niche of translation and localization companies and professionals by offering good performance, stability, and support for a wide gamut of formats. Users of memoQ can tell you wonders about the software, the company, and everything in between. Kilgray has just released MemoQ 8, codenamed Adriatic, and has announced a overhaul to its approach to software revisions. Until memoQ 2015, in fact, Kilgray released two version of its software each year: a “vanilla” version and an R2 one with incremental updates. Starting from MemoQ 8, however, Kilgray plans to release three or four updates every year named after seas. In a blog post, Kilgray teased their name could be Aegean, Alboran, Amundsen and Arabian.
The release of a new piece of software always begs the question “is this new version Worth consideration?” and “if I become an early adopter, am I at risk of running into unforeseen issues, like instability or regressions?” Those are important questions that everyone running a business that relies heavily on IT should ask herself.
Installation and first impressions
The installation process for MemoQ Adriatic is identical to that of previous versions: the installer is the same, and so is the wizard that appears once you run it for the first time. Just click Next until the installer tells you you’re good to run the program. The very first time I installed Adriatic on my main rig… It crashed miserably. I didn’t even get an error message or anything, but Windows Event Viewer showed an obscure error message. I contacted Kilgray’s support and they told me that this is a known bug that will be addressed in a future build and the only known workaround is installing version 7.8 (i.e., MemoQ 2015). I was a bit puzzled, since apparently I’m not the only one running into this issue and what they suggested isn’t a workaround at all. I then decided to install it on my MacBook Air. For this, I needed to create a Boot Camp partition and a clean installation of Windows. This time, everything went smoothly. Sure, I can’t take advantage of my desktop’s bigger screen and higher resolution, but it’s better than nothing. And besides, I’m doing this for science.
When I finally managed to get MemoQ 8 up and running, I noticed that everything looks the same. Yes, MemoQ 8 has the same interface introduced in memoQ 2014, with the ribbon bar in lieu of the older drop-down menu-based interface inspired by Office 2003.
I have to admit: when I fist saw this dashboard I shivered at the thought that this was yet another incremental update that doesn’t really improve on anything feature-wise. And I was right: MemoQ 8 will not blow you away with its increased performance or new functionality that will benefit every single user of the program.
To test this program’s performance I devised a small battery of tests regarding startup times, the importing of translation memories and term bases and creation of new projects. All tests were carried out on the same computer, a MacBook Air running the latest version of Windows 10 (build 14393 at the time of this writing). The specifications of the computer are as follows:
- Processor: Intel Core i5 4260U @ 1.4 GHz (dual core, hyperthreaded, 2.7 GHz boost)
- Memory: 4 GB DDR3 @ 1600 MHz (dual channel)
- Storage: PCIe-based SSD Apple SD0128F manufactured by Sandisk (46.6 GB partition)
We will be using Windows Performance Monitor to keep track of a few important metrics:
- General processor usage, expressed in percentage;
- Processor time of the application, expressed in percentage;
- Memory usage of the application;
In addition, we will be using a stopwatch (well, a stopwatch app on an iPhone, really) to keep track of how long each test lasts. We will be pitting MemoQ 8 against MemoQ 2013 R2 and Trados Studio 2017. In the case of MemoQ 8, we will analyze both the 32 bit and 64 bit versions for the sake of completeness.
This test is pretty straightforward: we measure how long it takes from double-clicking the MemoQ icon to having a usable application with all features ready to be used. I’d be the first to acknowledge that this metric alone is useless when assessing a program’s performance, but it can nonetheless be helpful for those who wish to run a CAT tool on a computer that’s not the fastest.
This test didn’t yield any significant surprises: even though MemoQ Adriatic was 1 second faster than MemoQ 2013 R2 and almost 1.5 than Trados Studio 2017, all tested applications were rather quick to start up
Importing a translation memory
This next test takes the files from the DGT translation memory (2004 Vol. 1) and evaluates how long the program takes to convert it into its native translation memory format. Since MemoQ 8 still doesn’t support importing multiple translation memories at once we’ll have to resort to using the process described here.
Now these results are interesting: despite MemoQ 8 being slightly slower than MemoQ 2013 R2, it still manage to completely obliterate Trados Studio 2017 by being almost twice as fast. Not to mention, MemoQ allows to import translation memories and term bases in the background, which is great, since it allows to work on other stuff while the resources are being imported. It should be noted, however, that translation memories and term bases being worked on are inaccessible until the import is finished.
Let’s also have a look at how Adriatic handles memory management compared to the other two programs. I know this graph doesn’t fit with the site’s overall visual theme, but I found that using the usual shades of red theme made figuring out the graph a lot harder than it ought to be. I apologize for that.
Please note that not all portion of this graph illustrates what happens during the importing process: data collection must be manually started in Performance Monitor, so there are a few seconds between the start of data harvesting and the beginning of the TMX import. One thing is clear, however: MemoQ seems to perform a great deal of the importing in RAM and dumps everything to the disk when it’s done creating the translation memory, as the constant rise in RAM usage seem to indicate. This can explain why MemoQ is much faster than Trados Studio 2017: the latter is constantly waiting for translation units to be written to disk. And since mass storage (including SSDs) is much slower than RAM, this results in a longer time required to complete the task, but with the advantage of keeping memory usage in check. Not that it matters, however, in a world where 4 GB of RAM or more is commonplace. I do, however, regret not keeping track of disk usage, for this would have been a useful metric to confirm or disprove this theory.
There is, however, one instance where the MemoQ approach is detrimental: importing huge translation memories (2 GB in size or more) with a 32-bit application. I recall importing the entire DGT translation memory once, which “weighs” about 12 GB, just to see MemoQ crash after three hours or so, because the application couldn’t access more than 4 GB of RAM. The importing went fine with MemoQ 64 bit, but it lasted 8 hours, so I’m not going to repeat that same mistake.
Overall, MemoQ 8 is slightly slower than MemoQ 2013 R2 and uses more RAM, which can be detrimental if you’re still running older hardware.
I don’t believe that analysing processor times by thread will yield surprises. As we’ve discussed in a past article, CAT tools do not benefit from Hyper Threading. Even more, Hyper Threading alone is actually detrimental to their performance. This suggested that CAT tools are generally single or lightly threaded application.
As I expected, this graph proves that importing a translation memory is mostly a single-threaded deal, meaning that the importing is performed by a single core in a computer. Only one thread (named Thread #2 in the 32-bit version and Thread #27 in 64 bit) is doing anything CPU-intensive, whereas most of the others are idling or not very much taxing for the processor.
Note: A thread usage of 100% does not mean that the processor is dedicating 100% of its time to that thread, but rather the thread is using 100% of the time the CPU devotes to that process. If the process itself is consuming 10% of CPU time and the thread is at 100%, it means it’s consuming 10% of the total CPU time.
CPU usage of a program can help determine which version will run better on older processors, as well as giving interesting insights for those who are on a newer platform as well. Generally speaking, a well-coded program that consumes a lot of CPU time leverages better a higher-clocked chip. Of course, there are exceptions, but this is a simple rule of thumb that generally applies.
As you can see, the 32 bit version of MemoQ Adriatic used less CPU time, which is helpful on older hardware, considering that it also uses less RAM (almost 100 MB less, in fact). However, this won’t be all that useful if your translation memories are big in size, as we discussed earlier. I’m actually surprised about the result given by MemoQ 2013, which I used to consider a lightweight CAT tool, but apparently this is not the case. But this is what happens when you can’t find reliable reviews for niche software like this.
Trados Studio 2017 fared slightly better than MemoQ Adriatic 64 bit, although not by a long shot.
This test was performed with this Wikipedia page saved as a PDF document and imported into the various CAT tools. I believe that this is the most inconclusive data of all tests carried out, because of how the three programs tested handle importing files: MemoQ 2013 creates the HTML preview first, whereas MemoQ 8 only does it at the very end of the process (when the user clicks Finish). Moreover, the readings for MemoQ 8 and 2013 R2 were taken just for the importing process, while those for Trados Studio for the whole project creation. The only important data we can gather from this is that Studio 2017 is much faster at creating new projects, really.
On average, MemoQ 8 used quite a lot more RAM than MemoQ 2013 R2 and Trados Studio, with the 64 bit version using more than twice as much as SDL’s CAT tool, which also proved a lot faster (Trados Studio’s time is for creating a whole new project, not just importing the PDF file). While it’s true that MemoQ 2013 uses about as much RAM as Trados does, it should be noted that it’s also much slower.
Now we shall analyse what threads are active during the PDF import. First up is MemoQ Adriatic 64 bit.
In this graph, the orange line (Thread #16) represents the thread the program uses to import the PDF file, whereas the blue one is the thread that does all the work during the HTML preview and project creation, proving once again that CAT tools do not use more than one thread.
A similar trend continues with the 32 bit version. The spike in the center represents thread usage during the import itself, whereas the one on the right side of the graph the finalisation of the project. I’m not sure why the program is only using one thread now instead of two, and I’m also not sure why Thread #19 isn’t getting 100% CPU time. I double-checked and, while there were a few threads doing very light work (about 2% CPU time), they don’t seem to be doing much at all.
And for the sake of completeness, here’s the thread usage for MemoQ 2013 and Trados Studio 2017.
Again, nothing too surprising to see here.
In this case, Adriatic 64 bit actually used less CPU time compared to its 32-bit counterpart. I’m not sure why this is, so if anyone has a theory, her suggestions are welcome. Both versions, however, used less than 30% of CPU time, suggesting that they could be used in a machine with generous amounts of RAM and a rather weak processor.
Again, MemoQ 2013 R2 proved to be the most resource-intensive of the bunch, whereas Studio 2017 used less than 32% of CPU time on average.
The blog post I linked to at the beginning of this article showcases a few additions that Kilgray has brought about to MemoQ 8. And if I can be completely honest, I fail to see how some of them could benefit me. Tracking changes in the source text? I generally do not deal with work in progress documents, and neither should most of the colleagues I know. It’s not a horrible thing to have by any stretch of the imagination, of course, but this is probably something that only corporate users will care about, not individual translators or small language service providers. And to make matters worse, tracking changes in the source text is only available for Word documents: all other file formats will need to be reimported manually. You can, however, keep track in the target text for any
Another addition I really don’t care much about is the change in how the Shift + F3 keyboard combination works. In previous versions, it would behave exactly how it does in Word (and even WordPad, for that matter), cycling through various types of capitalizations. In Adriatic, however, Kilgray introduced a drop-down box similar to that that pops up when Ctrl is pressed to insert a term base suggestion or a tag. Again, it’s nothing that’s bad to have, but not a deal-breaker either.
But it’s not all horrible: Kilgray actually included a few new functionalities that most people will find very useful, like filtering a specified percentage range of fuzzy matches and an improved QA system. I haven’t had the chance to see the QA system in action because I haven’t used Adriatic in a production environment (and I don’t plan to for the foreseeable future), but on paper it sounds impressive.
With most people still using Full HD monitors instead of 1440p or 4K ones, this isn’t an annoyance that will bother many, but it’s still worth to point out that resolution scaling in MemoQ is still broken, making its usage with a large TV (not that you should use one as your main monitor in the first place) or a high-resolution monitor a nightmare. Kilgray’s website claims that 125% scaling is not supported, but I still managed to get it to open at that value, but everything about its interface became a blurry mess. I swear, one of the reasons why I’m not getting a 4K monitor is that I don’t want to resort to an additional monitor just to be able to run MemoQ.
But one of my biggest gripes about MemoQ is that it still doesn’t allow to import multiple translation memories or term bases at once. Trados Studio 2007 did, and that program is 10 years old now. Come on, Kilgray, it’s a relatively simple modification, not rocket science!
Lastly, MemoQ Adriatic’s term base management is pretty much identical to what we’ve seen at least since version 2013. Wildcards and pipes are supported, but their usefulness is limited to isolating and agglutinating languages. You can still use a pipe (|) to mark the beginning of an optional part of the word and a wildcard to mark the variable part, but you still can’t define two different desinences (a term either has a desinence or it doesn’t in MemoQ).
Bottom line: Should you make the switch?
If I can be completely honest with you, no, you really shouldn’t. In my limited experience, I’ve seen that new versions of CAT tools are generally released with a slew of annoying bugs, like those that prevent you from running the program at all. Even if you’re paying your 125 € yearly upgrade plan, upgrading to Adriatic right away seems pointless to me, considering the notable regression I discussed. And even those who are just starting out as a freelance translator might want to hold onto their OmegaT or Virtaal for a bit, unless until the issues have been fleshed out in a future patch.
My advice is to wait until a R2 comes along, bringing about improvements in performance (although I don’t see how huge performance leaps could happen), stability and overall functionality. And could we please get a CAT where resolution scaling is not utterly broken, pretty please with cherries on top?