Editing existing translations and submitting changes back to the central server 4. Obtaining translations from the central server and rendering in-context (automatically or manually) 3. Once enabled, BE-COLA allows the following tasks to be performed by the user: 1. The following section graphically illustrates the BE-COLA architecture in detail. We also extend the functionality of the central server in this architecture by equip- ping it with a component to maintain TMs for different language pairs, a desirable feature in Translation Management Systems (TMS). The browser extension implements the UpLoD architecture, which acts as a proxy between the browser and the central server. However, in this architecture, clients (browsers) connect to the central server via a browser extension. The BE-COLA system resembles the Update-Log-Daemon (UpLoD) based client- server architecture. By keeping track of voters, users can be encouraged to vote for additional translations using ranking systems similar to those implemented in games. The server needs to keep track of the voters as well as the votes. The central server compares the new changes with the existing translations and includes it as an alternative translation.
Once editing has been performed, the new translation is sent back to the central server. If the user wishes to edit the existing translation, they can simply go to the in- context edit mode and edit the content. Considering the motivation factors related to crowdsourcing, a simple “thumbs up, thumbs down” voting is proposed over complex and confusing rating systems. In that page users can vote for any of the alternate translations. Furthermore, clicking on an alternative translation will take the user to a web page where the user can see all the alternative translations that are proposed for the selected segment. The users are given the opportunity to edit the current translation and/or to vote for any of the alternative translations. The votes for each translation will also be displayed next to their associated translation. When a user right clicks on a translated segment it can bring up a context menu where the current translation, along with the top 3 alternative translations, are displayed. Additionally, human intervention (mainly the opinions of expert monitors) is also essential to solve potential conflicts. However, in contrast to the existing UpLoD implementation, this voting mechanism is distributed to the entire community insuring a ‘many eyes’ philosophy. To assist in selecting the best translation for a given segment, a voting mechanism is proposed. When leveraging the TMs on the central server, a mechanism has to be built to choose the most appropriate translation of a given text segment, as per the UpLoD architecture. Furthermore, later on, MT systems can be trained from the TM data and these trained MT systems can feed back into the system to speed up the translation process and to translate content where TM matches are not entered. The central server can be scheduled to periodically leverage translations as the TMs grow. Additionally, a mechanism to monitor and uniquely identify browser extensions con- nected to the central server could be implemented, in order to prevent repeated misuse of the localisation service. Thus TMs get generated for specific locales, over time based on micro-crowdsourcing. Upon receiving translations from a client (browser extension), the central server stores all the information that it retrieves (the locale, the language pair, source string, target string, contextual clues such as XPath location, surrounding text, tags, URL, client identifier, client IP, text positions and text length) in a special TM.
Hovering the mouse pointer over a translation will display the original content as a pop-up (similar to Google’s web-based MT system). The browser extension keeps track of all translated content. Once the translation is completed, the extension sends the translated segment, original content and contextual information including the URL to the central sever. The extension allows in-context translation of the selected content segment in an editing environ- ment similar to Wikipedia. Similar to UpLoD, right clicking on selected text will bring up a contextual menu where a “Translate” sub-menu can be found. That is, it allows a selected text segment to be translated into the user’s preferred locale. The browser extension also facilitates the in-context translation by the viewer of the source content. For replacing the original text with target strings, techniques such as Regular Expression matching and XPath queries may be utilised. Finally, the content will be rendered in the browser. will replace the actual downloaded content (source strings).
0 kommentar(er)
