2.4 Test framework
2.4.3 TM systems
2.4.3.1 Products and versions
The TM systems tested are listed below. Their version number(s) are in square brackets, their manufacturers in parentheses.
Across Standalone Personal Edition [4.00] (Across) Déjà Vu X Professional [7.5.303] (Atril)
Heartsome Translation Studio Ultimate [7.0.6 2008-09-12S] (Heartsome) memoQ Corporate [3.2.17] (Kilgray)
MultiTrans [4.3.0.84] (MultiCorpora)
SDL Trados 2007 Freelance [8.2.0.835] and SDL Trados Studio [9.1.0.0] (SDL) Transit XV Professional [3.1 SP22 631] and Transit NXT [4.0.0.672.3] (STAR) Wordfast [5.53] and Wordfast 6 [2.2.0.4] (Wordfast)
The selection had to balance two separate requirements: on the one hand, the selection had to be as representative as possible of the available TM systems; on the other hand, the selection had to be limited given the impossibility of investigating all available TM systems. The selection was made according to two criteria:
Including popular TM systems.
29With the exception of Heartsome, which requires a conversion into the DOCX format, rst introduced
Including (at the time of selection) less common TM systems that oered interest- ing features or the potential to gain importance admittedly a highly subjective judgement.
Popular TM systems at the time of selection (end of 2007) were and are Across, Déjà Vu, MultiTrans, SDL Trados, Transit and Wordfast. Less common were Heartsome Studio and memoQ. In the meantime, memoQ has proven to be innovative and has stood its ground. Heartsome was to some extent a precursor of the ongoing trend of reliance on open standards, see 1.4.5.
If carried out today, the selection would probably be dierent, see e.g. the selection made by Keller (2011). This limitation applies to all comparative tests: after some time, tested applications may be taken o the market, become obsolete, merge, etc. The market is anything but static: magazines such as MultiLingual and MDÜ regularly publish articles on new applications, new releases etc., see e.g. García and Stevenson (2010), Geldbach (2010c) and Zerfaÿ (2010).
2.4.3.2 General TM system settings
The TM system settings are important for the repeatability of tests. The standard settings were sometimes modied in order to fully exploit the functionalities oered by the TM systems. However, customization was limited to activating/deactivating existing functions. Some TM systems allow for deeper customization, e.g. memoQ can be enriched with regular expressions, see also 2.4.4.1. Despite its usefulness, this feature goes beyond the presumable knowledge of many users an assumption conrmed by (Zerfaÿ, 2010, 17) and the results do not depend on the TM system, but solely on the user's skills. Any adaptation of this kind would bias the comparability of the results and was consequently avoided. Nonetheless, customizability is an aspect of the evaluation, see 2.2.3, and the customizability options applied when testing placeable or localizable elements will be discussed in the section Customizability of each chapter.30
The settings that were relevant for all tests are presented below. Further settings, which were relevant only for some test subsets, are described in the section TM system settings of each chapter.
Minimum match value
The lowest possible minimum match value was set for all TM systems. This setting was necessary in order to evaluate as many suggestions as possible, but is generally not advisable during translation because it causes unhelpful fuzzy matches to be suggested. The following list includes all TM systems tested along with their default minimum match value, their lowest possible minimum match value, and the path to this setting.
30The same considerations and methodology can be found in Makoushina (2008), albeit the test object
Across: default 50%; min. 50%; Tools > Profile settings > crossDesk > crossTank > crossTank search settings > Minimum matching degree. Déjà Vu: default 75%; min. 30%; Tools > Options > General > Database
Lookup > Minimum Score.
Heartsome: default 70%; min. 35%; Options > Set Minimum Match Per- centage.
memoQ: default 30%31; min. 0%; Tools > Options > TM defaults, user
info > Thresholds and penalties > Match threshold.
MultiTrans: default 75%; min. 0%; Options > Fuzzy Matching > Fuzzy Factor.
SDL Trados 2007: default 70%; min. 30%; Options > Translation Memory Options > General > Minimum match value.
SDL Trados Studio: default 70%; min. 30%; Tools > Options > Language Pairs > All Language Pairs > Translation Memory and Automated Translation > Search > Translation > Minimum match value.
Transit XV: default 70%; min. 0%; Options > Profile > Settings > Fuzzy Index > Min. quality for fuzzy matches (%).
Transit NXT: default 70%; min. 0%; User preference > Dual Fuzzy > Source language > Minimum quality (%).
Wordfast 5: default 75%; min. 50%; Setup > General > Fuzzy threshold. Wordfast 6: default 75%; min. 40%; Edit > Preferences > Translation
Memory > Fuzzy Match Threshold in (%). Other settings
Across: under Tools > System Settings > General > crossTank, the default option Use Rich Translation Memory is activated. Rich translation memory (rich TM) means that formatting, inline objects and other characters (control characters, special characters and spaces) are saved in crossTank, Across (2009a). The conguration is carried out under Tools > Profile Settings > crossTank > Storing, where the options Store format ranges, Store inline objects and Store special characters are active.
Déjà Vu: the options AutoAssemble and AutoPropagate under Tools > Options > Environment are active. If no match can be found for the current source segment, Au- toAssemble assembles the translation from several target language fragments coming from
dictionaries, terminology databases or translation memories. This functionality depends also on other settings, see (ATRIL, 2003, 97).
If a source segment repeats in a translation project, after conrming its translation when it rst occurs, AutoPropagate automatically inserts the translation as a 100% match into all other occurrences. This feature is relevant because of its interaction with automatic adaptations.
memoQ: when creating a translation memory, the option Store Formatting is ac- tive. Formatting information as e.g. font attributes (bold, italics etc.) as well as formatting tags are saved in the translation memory.