Jan Van den Bergh is a post-doctoral researcher and research assistant at Hasselt University and member of the HCI lab of the iMinds research institute Expertise Centre for Digital Media.
His research is situated in user-centred engineering of context-aware, mobile or collaborative systems. His recent research was often focused on how software can support workflow for knowledge workers and/or end users in specific domains. He obtained a PhD in computer science (human computer interaction) from Hasselt University in 2006 and a M.Sc.
He co-organized several scientific workshops and served as PC member for several conferences and workshops and is observing member of the IFIP workgroup 13.2 on User-Centred Systems Design
Jan Van den Bergh
Recommendations for Translation Environments to Improve Translators workflows
Language professionals play an important role in an increasingly multi-lingual society, where people commonly do not sufficiently understand all languages used in their environment. While there are many translation environment tools (TEnTs) available to support translators in their tasks, there is evidence that these tools are not used to their full potential. Within the context of a broad research project we investigated the current tools and work practices of language professionals to enable personalization of the user interfaces of translation environments and improve translators’ workflows. The project also targets improvements to machine translation, confidence estimation of machine translation output, terminology extraction from comparable corpora and speech recognition accuracy.
We used complementary research methods in our study, e.g. a survey among language professionals, semi-structured interviews with 5 local companies involved in the translation process (technology) and 9 contextual inquiries with both in-house and freelance translators and revisers.
The survey was completed by 181 respondents (119 female, 62 male), 75% from Europe. Slightly more than 75% use TEnTs. While 50% of the users prefer to pre-translate the source text with the help of translation memory, only 10% use machine translation. The usage frequency of specific features varies according to the user’s role in the translation process. Most respondents indicated they did not receive support to enhance their knowledge of TEnTs.
The semi-structured interviews focused on the workflows and the roles involved in the translation process used in the companies of the interviewees as well as specific desires for TEnT-related research. Interviewees reported a need for flexible user interface designs with customization options that allows users to adapt the tools to their individual workflows. Live previews or WYSIWYG are desirable features within the translation editor. These findings are in line with the results of our survey and that of Lagoudaki (2009).
Contextual inquiries revealed more information on the work practices. Based on this information we identified eight relevant scales to typify the users and their experience with TEnTs: interaction techniques, media used during translation, software knowledge, received training, support availability, number of active languages, level of customization, and professional environment. We created generalized workflows, and we summarized the key insights using two personas.
Based on the reported research we propose a number of recommendations that could positively impact translators’ workflows. These recommendations are in line with but go beyond what has been reported in state of the art:
Improve efficiency of TEnTs by offering keyboard shortcuts for all (frequently used) functions, which was already recommended by Lagoudaki (2009) and confirmed in our research. The interoperability of files and projects between TEnTs should be optimised as both translators and agencies use more than one tool during their projects.
Improve effectiveness of translation environments. The latter includes not only the TEnT but also all other digital and physical resources used to make the translation. Effectiveness could be reached by better integrating online and/or physical resources (translation memory databases, dictionaries, reference materials) into the TENTs. TEnTs could be improved regarding the way in which results of machine translation are presented, as recommended by Green et al. (2013). Changing feedback given by TENTs during translation to include normalized BLEU scores or social messages, as suggested by Tsai and Wang (2015) can also improve effectiveness. Providing a visual context for the translation may also be beneficial for the localization of user interfaces as noted by Leiva and Alabau (2014). Similarly, Leblanc (2013) noted limited availability of contextual information as an issue for TEnTs.
Improve usability of translation environments. Both the survey and the field research revealed usability issues in the current TEnTs. Specific actions can incorporate the recommendations of Lagoudaki (2009), which include many features already available in word processors.
Give translators more control over their translation environment. Smart translation environments should assist the translator, but not take over. They should suggest improvements, but not force them upon the user. Specific activities in the translator workflow may be supported through a specific combination of tools and/or configuration of the TEnT. Green et al. (2013) recommended that tools should not try to predict the specific activity. Activity-based computing environments, such as cAM (Houben et al., 2013), may ease manual switching between different tool combinations or configurations between activities.
We believe these recommendations can assist the TEnT developers in making decisions on the evolution of their software. The insights from our studies and the recommendations can also be used by researchers and language service providers to improve translation environments beyond the TEnTs.