Quality in Translation
From Ott09 Wiki
- 1 Ariel's notes
- 2 John's notes
- 2.1 Measures of quality
- 2.2 Glossary
- 2.3 Quality of translation memory
- 2.4 Segmentation
- 2.5 Quality should be tested early and automatically
- 2.6 Some things can be automated easily (like spell, glossary, check for spacing and formatting); some things not (like grammar)
- 2.7 Workflow
- 2.8 Frequency of strings
- 2.9 Reputation system
- 2.10 View "context"
- 2.11 Being able to "preview" translations immediately
- 2.12 Central repository for terminology
- 2.13 Potential co-operation with National Standards/Terminology Boards
- 2.14 Automate the process as much as possible
What is quality?
Things that should be checked:
- Style, mistranslations, what other issues do we worry about?
- (We had someone specify that you must translate into your native language. What do we think about that?)
- Correct grammar, spelling, punctuation.
- Correct use of placeholders (variable + grammatical indicator)?
- Maintainenance of inline formatting elements or other elements that should not be translated
- Translated string length should approximate the source text for some uses (e.g. text strings from software)
- Awareness of the audience when translating (what is the audience used to, what do they expect?)
- Style of the text (1): do we want a freer translation that conveys the meaning, or a translation that adheres as closely as possible to the structure of the original text? (Literature vs software interface for example)
- Style of the text (2): uniformity of style, in the case of multiple translators (collaborative translation efforts)
- Good and accurate glossary
- Quality of translation memory, information about its source may help us to rate it
- How cu lturally dependent concepts are rendered in the target language
- Central repository for terminology? (But this body may mandate translations of terms that are not used in practice and that seem laughable to the reader) Depends if they are prescriptive or descriptive, how successful prescriptive efforts are, etc.
- Indication of the text difficulty?
- If the file with the translated text also has the translation memory and the glossary, that's easier to manage (for coordinated translation efforts)
- Workflow is important
- Which strings are viewed more often (so you translate those first, in software string translation)
- Means for the translator to view the string for translation in context (screenshot?) for software string translation; be able to see translations applied to a web page (software-generated or with content as well)
- Means for the translator to view the translated string in context as well
- Translator reputation system: do they meet their deadlines, quantity produced, quality of translations
- Make sure reviewer only has to check updates to files (diff, thanks :-) )
For example, mark messages as translated, or translated and reviewed.
- When do you do a quality check? Hopefully you do some of these checks early on so that systematic errors can be caught before the translator has invested a lot of time on the work.
- Community dynamics?
- Translators can choose what they translate; how does this relate to the quality question?
- What about workflow issues?
- Mechanism for feedback to translators after review, so they become better translators? Track changes for example?
- Edit the initial content to make it easier to translate (breaking up long text for example)?
- Are their automatic grammar checkers? (There is some sort of basic grammar checker for OpenOffice: Irish has checking for case and initial mutation, there isn't anything good for English, there is nothing good for Chinese)
- Many translation memories are not open (and access costs $$)
- There are many factors for measuring quality, hard to determine a scale of measurement
- Workflow and tools both may play an important part in supporting better quality translations
- Workflow should give feedback to translators?
- Quality should be tested early and automatically
- Reputation systems for translation memories and for translators may help us generate better quality
- We need to develop more glossaries and terminology banks
Other ideas for discussion
- Terminology standards
- Reputation systems
- Quality testing automatically: how do we do this?
- Automated grammar checking
- Workflow: what does it look like?
- Previewing translated text and viewing original text in context
- Data formats for all of this metadata
Measures of quality
- Industry translations at the top level; consistent
- Need effective ways to measure it
- Correct glossary & terminology
- Correct grammar (like language tool on openoffice)
- Correct spelling
- Correct placeables use of "variables" (like %s)
- Maintain inline elements and formatting (like ... )
- Translation length (in certain environments; i.e. subtitles, mobile)
* language dependent
- Show / enforce translations
- Terminology banks
Quality of translation memory
- What is a translation memory (for people who don't know)
- More than 50% of EU translations come from memories
- Quality of a translation memory important (a bad translation memory will lead in bad translations)
- Need information about it (i.e. origin)
- being able to cut a paragraph into sentences
- so that one can look them up in translation memories
Quality should be tested early and automatically
- If done too late: wastes time
- Can the program "do it" (ask you why didn't you use the glossary, kept existing formatting etc)
- Any test that can be automated ... should be automated and done by the product
Some things can be automated easily (like spell, glossary, check for spacing and formatting); some things not (like grammar)
- Reviewer needs to know which messages are new
- Workflow information should be embedded in the files
- Should be interactive to a certain degree (reviewer's feedback reaching the translator and vice versa)
- Should help improve quality in the long term
- Should be adjustable
Frequency of strings
- The most viewed strings should be translated first
- Systems like MediaWiki have this information
- Publishers to control translators
- Factors are unclear; "quality / quantity / speed / reliability"
Being able to "preview" translations immediately
- For instance when translating webpages, a preview of the translated webpage
- Important in UI as well
- For instance viewing changes instantaneously is easy with GTK+, hard with Openoffice