6 QUALITATIVE PHASE: DATA ANALYSIS
6.4 Translation
6.4.2 TM tools
result is that, as suggested by Bowker in section 2.7.1, TMs inevitably become
‘dirty’, i. e. filled with inconsistencies, and require maintenance to clean them up.
F (QA specialist) said that cleaning TMs retrospectively is “much too expensive to do”, aside from the difficulty of editing an active memory that is being “added to and leveraged from on an hourly basis”. K said that to keep TMs in perfect shape is “a huge, huge effort”.
Seven interviewees mentioned QA tools that they use for TM maintenance: three (G, H, and M) use ApSIC Xbench, two use QA Distiller (K and M), and two (F and I) use internal proprietary tools. Other than assiduous maintenance of TMs, some interviewees suggested that inconsistencies can be minimised by controlling the translation process. I (PM) said that while “ten glossaries, no penalties, 15 TMs”
may result in a lot of leverage, she and interviewee K both stressed the benefit of small, targeted TMs that provide a low number of relevant suggestions, in effect prioritising precision over recall (see Bowker, section 2.7.1).
6.4.2 TM tools
Interviewees’ answers to IQ3 and IQ7 revealed various frustrations with TM tools.
They cited dissatisfaction with the evolution of TM tool functionality (8 interviewees), with matching and segmentation (5), and with how tools deal with tags in ST (7).
Evolution of TM tools
Interviewees felt that there has been a lack of useful development of TM tool functionality (see section 2.2 on the history of TM tools) and that a lot more could be done within the tools to minimise inconsistency. This can be seen by the number of interviewees who discussed their requirement for separate QA tools.
Instead, the focus from developers appears to be on workflow and technical improvement of the TM system (see discussion of Lagoudaki 2008 in section 2.5).
Amongst the interviewees, translators and non‐translators were dissatisfied by tool development in different ways. G (QA specialist), for example, felt that rather than changing how the tool works with data, developers are changing the workflow and he says “actually I haven't seen that much development” (see section 6.7). This was typical of the comments of non‐translators. Most of the SDL Trados users cited similar reasons for not upgrading to the more recent Studio iteration of the software. K (COO) said that developers may have “put bells and whistles around it (the software)”, but compared to the older TWB functions are “pretty much the exact same”. She feels that the TM tool has become a translation management system, “whereas it should really, truly, focus back on consistency, language quality, and really incorporating and integrating all of those other functions”.
In section 2.5 we also discussed how TM technology has been said to serve industry goals rather than users’ needs. This theme recurred particularly when speaking to translators in the current interviews. In answering IQ7, C (translator) recalled when she used to work for a TM tool developer. She said she would
“give the developers feedback and comments” but that “they can't really see it…
they haven't translated ever, so they don't know the problems you encounter or the things you would like to see”. As a result, she finds that TM tools show too much focus on “superfluous” design aspects, such as colours and options, whereas the translation interface leaves little room for context. She says that “if translators had the knowledge and the capacity to develop the tools as they use them, no doubt they would be better”.
B spoke about the problem of interoperability as another reason for not upgrading or changing TM tools. She cited a loss of leverage when moving between tools, and the effect that this can have on consistency, explaining that
tool, even then you can have a loss”, and developers “wouldn't acknowledge [it], but it happens”.
Matching and segmentation
In section 6.3 the interviewees agreed that ST segments with minor inconsistencies often evince further inconsistency on the TT side, as seen in Chapter 5. Two ways of preventing such TT inconsistencies would be by controlling the ST or by ignoring certain minor inconsistencies in the TM tool. K, in response to IQ4, said that if ST segments are “even one space different, one comma different, it (the TM tool) won't pick them up as being identical”, rather
“it will still show up as a different segment”. However, eight interviewees (including K) said that many 100% match suggestions are erroneous. E (translator) said that, when a glossary has been updated, the TM may have the
“old style and old terms, so you have to review the 100%. You can't trust 100%”.
K concurred, “You might have stuff in the TM, but really, you know, they're not 100% matches”. The interviewees also find that, in the case of fuzzy matching, translators may “over‐correct” (F, also mentioned by G, I, and J), or spend more time editing a fuzzy match than it would take to translate from scratch. This view concurs with research reviewed in section 2.6.2 that found that, below a certain fuzzy match threshold, TM matches cease to be beneficial.
Several interviewees also cited problems with segmentation. L (workflow manager) finds segmentation limiting for translators, removing flexibility and context. D (translator) and M (software localisation engineer) said that segmentation rules are designed for English and are not always appropriate for other languages. D has had problems with French language segmentation and wishes that the software would “recognise more signs or ways to segment the text grammatically depending on the language.” She said that French text may have an “unbreakable space before exclamation marks, question marks”, which often leads to a segment containing “the mark alone, without the rest of the
sentence”. C (translator) has had difficulties with incorrect segmentation at URLs, tags, and at other types of non sentence‐ending punctuation (see section 6.3.4).
Tags and format issues
In Chapter 5 we found a high rate of category 1 tag inconsistencies in each of the four TMs (particularly in TM D). Six interviewees said that they found dealing with tags in ST was a problem for TM tools. B (translator) said that “sometimes due to difference[s] in tags you get a different leverage” and that this is exacerbated by tags being handled differently by different tools. M feels that it is too easy to modify tags in TMs and C has had problems with tools disallowing the removal of unnecessary tags. She has sometimes been unable to convert the files back to the original format from her TM tool and as a result, “the only way you can sort that out is in the final document”. This means that she has to leave unused tags at the end of segments and leave instructions for engineers to remove them, which she finds unsatisfactory.
L also said that tags had caused inconsistencies in his TT when STs were incorrectly segmented at the tag, separating a single segment into two segments that “you can't reconcile… unless you can merge the two segments”. B finds that when moving between TM programs, “every tool handles tags differently” (see sections 6.4.2 and 6.6). M said that his solution to problems with tags is to remove “unnecessary formatting” from the text if the client is agreeable. For him, a line break or hard return in the ST provides a more difficult problem to resolve, and has a “great impact because you’re not gonna get any leverage”. H complained that, although the latest version of MemoQ allows translators to merge segments “even if there's a hard return, which is one of the biggest problems with Trados and all the other tools”, translators are unused to being able to merge the segments, so they still tend not to.