Translator's Shack

Translator's Shack is a collection of links, news, reviews and opinions about translation technologies. It's edited and updated by Roberto Savelli, an English to Italian translator, project manager and company owner of Albatros Soluzioni Linguistiche, a team of English-Italian translators, which hosts and supports this blog.

The Life as a PM category, managed by Gabriella Ascari, contains topics that are less technical in nature, but which we're sure will be appreciated by owners of small translation businesses and freelancers.

Here are links to my pages on some social networks:

Dynamic Translation Memory: Using Statistical Machine Translation to improve Translation Memory Fuzzy Matches

I came across this interesting paper by Ergun Biçici 1 and Marc Dymetman. In short:

The paper proposes to perform this automation in the following way: a phrase-based Statistical Machine Translation (SMT) system (trained on a bilingual corpus in the same domain as the TM) is combined with the TM fuzzy match, by extracting from the fuzzy-match a large (possibly gapped) bi-phrase that is dynamically added to the usual set of “static” bi-phrases used for decoding the source.

Here is the download link

Across Systems Releases Language Server v5 SP1

Across Systems announced Language Server v5 SP1. In brief, here are some of the new features:

Web-based Translation Workflows
Crowdsourcing
Integration of Machine Translation
Authoring Assistance

See the full press release here.

Importing the Microsoft Terminology Collection

The latest Tool Kit contains a nice description and details about importing the newly-available

Microsoft Terminology Collection into the translation environment of your choice. If your tool does not support the TBX format, however, you will have to transform the data into the proper format (e.g. CSV) before importing it.

The Tool Kit suggests using the excellent XBench for importing the TBX terminology file and exporting it into a comma-separated file. It also warns that XBench drops the “definition” field which, in my opinion, contains very useful context information. So in this case I’d say XBench is not the way to go.

By digging into the memoQ user discussion formum, I found this useful tidbit of information by Denis Hay:

True, we don’t have official support for TBX yet, but just add ".xml" to your file, open in Excel 2003 or 2007 and save as Unicode text. You will easily be able to import that into any memoQ termbase, picking only those columns you want.

Excellent. This should solve the problem and make the TBX easily accessible even if your favorite translation tool does not support this format natively (as is currently the case with memoQ).

Another solution that I tried and found to be working flawlessly is using Wordfast Pro, which supports TBX out of the box and allows you to export an imported glossary to CSV format. Wordfast Pro is available in a free trial version, which has some limitations. I’m not sure if the free version will allow to import and then export the whole Microsoft glossary, but my guess is it will.

CopyFlow Gold for InDesign CS5 5.2 released

CopyFlow Gold allows to export formatted text out of InDesign or QuarkXPress documents to a computerized translation system, and then batch import the translated text back into its original page location — all while preserving the typographic formatting. CopyFlow Gold for Adobe Illustrator is also available.

According to the latest post on the developer’s blog, CopyFlow Gold for InDesign CS5 has received an update.

some pictures from memoQfest 2010

Jeromobot: clearly Budapest is not a safe place for you yet. Now we know why you weren’t there…
		memoQ users among (other) kings

memoQfest 2010 - TM Repository

Moravia are the first adopters of the TM Repository.

The main problem with the TM repository so far was: “what is it, and what is it intended for?”

According to statistics, 80% of new software projects fail if there is no clear vision about the project. The TM Repository project has been put on hold for a certain period, but the company has resumed development now that there is such a vision.

Problem statements

TMs can be scattered in hundreds of tiny files.

Or, in a “big mama” type of approach, the TMs are managed by using filters and metadata. A project TM is a TM that only contains the segments (e.g. from a “big mama”) that are relevant to a specific project.

Including the reviewing process into the technology is essential. Some reviewers will refuse to use a translation environment to review a text (this has been solved by memoQ’s RTF columnar export).

TM contents need regular cleanup. What needs to be deleted/fixed: outdated entries, bad translations, “other” garbage.

In technical documentation, there is a lot of terminology that refers to concepts that get deprecated.

Cross-tool TMX transfer does not give good enough leverage. Attributes specified in one tool will often not be transferred to a different tool.

Context information: how can we transfer the context information (saved as hashes in Trados Studio TMs and as clear text in memoQ)? In short, we can’t. The hashing algorithm used in Idiom is patented. In other cases, it’s kept secret.

Features and workflows

Single database with TMX import/export. At its current stage TM Repository can only receive TMX files.
It accepts any sort of metadata (client, project number, dates, etc.), no matter which tool was used to create the TMX.
The TM is extracted from TM Repository and can be modified in any translation tool. The TM then needs to get back into the TM Repository, which offers complex features for comparing the new TM to the current contents of the repository.
You can import TMs containing one language pair and merge them into a multi-pair TM, then split it again in sub-pairs.
Version control and rollbacks: entry history is included for every translation unit. If you make a mistake, you can always roll back to the previous version
TM Repository is tool-agnostic. Every attribute is converted to an attribute system that is specified by the user for the TM Repository. Preconfigured mapping templates are available.
Full-text (not fuzzy!) concordance search
Search and replace in the repository

Next steps

A beta version is available now
A connector to memoQ Server is in the works. It will make import/exports from/to memoQ transparent through an integration API.

Specifications

Back-end database application with no direct connection to translation management systems (for now). Requires a dedicated server, as it’s resource-intensive.

In its current state, TM Repository will not allow you to upload a translatable project, run a fuzzy-match analysis, and extract the relevant translation units from the repository, regardless of the metadata. From my personal point of view this is a real show-stopper. I cannot see real potential in this product for a small LSP if it does not offer this kind of functionality.

memoQfest 2010 – How to be successful in the translation industry

Isabella Moore from Comtec gave a very interesting presentation this morning here at memoQfest in Budapest. I am posting the four final points that summarize the issues she discussed.

Communicate results to staff: this will improve the relationship between the company members at all levels (from management to employees) because it gives everyone a good feedback about the overall performance. Interestingly, this aspect is perceived as being very important by employees.

You can’t do everything yourself: learn to delegate and start doing it now.

Join networks, but be selective: some of them can be a waste of time and resources (this includes online networks, in my opinion)

Consider selling to the public sector: A considerable chunk of Comtek’s turnover comes from the public sector. Of course, approaching this sector requires specific strategies that are different from those that apply to the private sector.

memoQfest 2010 – Q&A with Kilgray management

After an in-depth explanation of the “do not press this button” button (short answer: you can press it if you find it in a software program but you shouldn’t press it if you are in a steam bath), a Q&A session started with Kilgray management.

Training opportunities and initiatives will be expanded. Some of the information about trainers is already on Kilgray’s website. Training is seen as a great opportunity for expanding memoQ’s user base.

Suggestion to allow translators to register as memoQ users in order to reach new clients, i.e. memoQ translator marketplace. There are no plans for such a program at this point.

memoQ localized interface and manuals will be available in many more languages very soon.

One recurring request is to have a stripped-down “server” version that would allow two translators to share a resources. The answer may be to adapt the existing offering rather than developing a whole new product.

Another requested feature is the ability to allow the insertion of short translated texts to a TM without going through the procedure of opening an existing project or starting a new one. There are no plans to add such a feature for the time being.

Final words of praise and a warning from one user: Kilgray, please do not lose your personality and accessibility and become a “corporation”.

The incredible shrinking delivery times

It’s nothing new to note that delivery times have been increasingly shrinking over the past 10 years. Everyone who’s in the business knows this is how things are now, and that’s that.

But what are the actual reasons for the ever closer due dates?

Is it always the fault of the unreasonable end client who will not take no for an answer? Is it the fault of the large LSP between yourselves (well, at least ourselves) and the end client? And why is it their responsibility? Have they lost their bargaining power entirely and simply nod and agree to any conditions? Or is it because they find it hard to organize a workflow which involves translator, reviewer, DTP editor, and translator again? Or is it perhaps because they are under the desktop publishers’ thumb and have no choice but to sacrifice the translator’s time?

I ask myself these questions daily, as I incessantly negotiate for extra time with my clients, but the answer can only be a guesswork.

My first explanation is that the economy is pushing and prodding end clients to issue their material earlier than reasonably possible. It is plausible. The economy has changed, recently for the worse, but even before that, communications have become so fast that a company simply has to keep up, or else.

My second explanation is that large LSPs are backing up and are no longer in a position to negotiate with a firm hand. I don’t think this was brought on by the recent economic crisis. I believe in fact it started earlier, with the gradual merge and consolidation of the major global language service providers. As a result, the smaller LSPs who are still in business are no longer prepared to require certain (and previously perfectly reasonable) conditions for fear their clients will go somewhere else. How else can one explain the fact that negotiation of delivery dates is not as part of the larger LSPs’ routine as it is part of mine, as I often have to urge them to ask for extensions as if they hadn’t even thought about it.

My third explanation is again a platitude, and that is that speed has become far more important than quality, and everyone is happy to live with that. Speed makes languages grow more similar to one another through previously unconscionable loans and it makes details go from paramount to expendable. Speed is preventing translators (and reviewers at that) from exercising their mind and their knowledge of the source and target language; instead, it forces them to resort to words and expressions that have grown to form their source database for this or that client to achieve speed. And the tighter the deadline the more this source database is pillaged and abused.

I have no particular qualms of conscience about the way things are going and I’m not a purist or perfectionist who still clings to a romantic view of translation. I do care about my mind, though, and I know that not exercising languages, even your own, equals to forgetting them. And this simply won’t do.

memoQfest 2010 – “ask the Geeks” Q&A session

New feature for memoQ 4.2 or later: project archival. Backup and restore functionality for the project including TM and TB. Useful for moving projects to a new PC and keep two PCs in sync.

Speed degradation on network drives: it’s a problem created by the underlying input/output system.

More on archiving and paths. Suggestion to make relative path as the default. This suggestion is being considered.

In 4.2 a brand-new aligner is available. Interface has been reworked, improvements with multiple documents. Segments are editable.

Problems in handling Chinese/Japanese content:The focus from the input method editor shifts away from the main window. Developers are working to solve the problem. It appears to be a rather complex bug. There is also an alleged problem with fonts since version 4.0. If you copy Japanese text from the source to the target and the target uses a Latin-based font, you may get useless squares. However, most users translating from/to Japanese do not have a problem with this.

Dragon Naturally Speaking support: support in the 4.0 version has problems. These should be solved in version 4.2.

Feature request: hide the mouse pointer when the user starts typing. This is probably already supported by your mouse driver.

Plans to have a clone project feature for local projects.

Feature request: search for tags and/or filter for tags. It’s being considered.

Request: make the metadata from the term base visible in the translation environment. In all probability, qTerm will be addressing this issue.

Hunspell problem with the Rumanian dictionary. Perhaps the best solution is to replace the default dictionary with a new one.

Ctrl-shift-B – Ctrl-shift-N keyboard shortcut allows to move the selection to the left/right.

Request: if the source term is in small caps, insert it as small caps even if it’s capitalized in the target term base.

Fragment-assembling: often short segments from the TM take precedence over the same term from the term base. Sometimes the result is that a term is inserted with wrong capitalization. Perhaps this issue will be solved in qTerm. qTerm will also offer filtering capabilities.

Inverting the “direction” of translation memories. Version 4.5 will introduce features that will make this “problem” obsolete.

Inline tags: if you have translated a project containing bilingual files that use the “old” (inserted by F8) type of tags, and then receive a new version that use the “new” version tags (F9), your match rates will decrease because the tags are not substituted on-the-fly. This is something the developers are working on.

Request: LAN access to term bases and translation memories without using memoQ server: this is recognized as an important feature, but it’s not going to happen. Tiny translator groups still have to purchase the server version if they want to share resources.

Request: add a keyboard shortcut to add a term as untranslatable. This is being considered as an addition to future versions.