Start a new topic

Pretranslation with a preset % of matches.

Well, maybe it's a silly question but, how do you perform a pretranslation with a preset % of matches? For example, to populate those segments that matches in a given % just to postedit it?


Hello, Victor,


TM FUZZY MATCH PRETRANSLATION


A similar question asked today on ProZ. Current answer here: https://www.proz.com/forum/cafetran_support/330244-new_to_cafe_tran.html


With the introduction of the new states [AI translation (MT-TM)] and grid symbols, CafeTran could be made to insert the “Preliminary memory matching” result with a separate action (Translate > Insert all Preliminary matching).


Why I suggest a separate action or to leave this as optional? Because I personally much prefer CafeTran’s own take on the matter. By the time you reach a specific segment, which would otherwise have been populated by a Preliminary matching result, the Automatic TM workflow option ensures more up-to-date matches that you might create while working on the translation will get a chance to be inserted instead. Performing the matching according the TM workflow option (or Translate menu action), but leaving the segment unpopulated until you actually visit it is a better way of proceeding, although I do understand the need for running actual pretranslation (fuzzy/autoassembling matches inserted in the segments), especially if a project is to be sent to someone else (for example in a Project manager/agency scenario). In that case, it does make more sense to perform such an option, and I’m OK that these states are not stored for external projects.


MT PRETRANSLATION

If someone needs to create an MT pretranslation instead (or mix it with an Exact matches approach), CafeTran already offers that possibility. Create TMX memory (right click inside the MT engine pane), then Translate > insert all exact matches, with only that TM resource loaded. In that case, it would still bare the [AI translation (MT-TM) symbol, right? 

 

This added action would add the possibility to: Insert exact matches (already possible) from a TM, plus the fuzzy matches according to the Preferences > Memory threshold settings, then load an MT TMX to populate the rest of the segments. Only problem with that is that the segments won’t get the MT status, they would get the TM match status instead. Two solutions: either just use the “Automatic transfer to target segments” option for an MT engine (this would work for an individual translator using the general MT engines CafeTran offers connectors to) or CafeTran could mark the TMX memories created out of an MT engine (possibly based on a specific TMX property that can also be manually assigned) with the MT symbol instead. Again, this could be useful in a Project manager/Agency scenario using either the standard MT engines provided by CafeTran, of custom/separate ones, if these can produce a TMX outside of CafeTran.

 

Igor, what do you think of that suggestion? I think it’s worth investigating if these two actions could make CafeTran better suited from an agency/PM perspective as well, or those that absolutely want a lesser “pretranslation” feature than the currently available one.

 

One question though, because interoperability works both ways. While I understand states/statuses are not necessarily kept for external projects, what happens if a CafeTran XLIFF file is translated in another tool? Do these features present a compatibility risk? Are any already marked states preserved or altered if the same XLIFF is reopened in CafeTran after the translation in another tool?


Jean

Thanks Jean! I will come back once I make some tests with your explanation.

The interoperability cannot precede usefulness, comfort, speed and progress. Pre-filling segments with fuzzy matches seems like both a waste of time and being error-prone due to tag differences (e.g. export without a QA tag check would almost certainly fail). The user usually needs to check and edit such fuzzy matches.  I guess that more often than not, they would need to rewrite the translation deleting a fuzzy match. CafeTran's implementation lets you control the fuzzy match insertion in real time and decide what action to take - insert machine translation, exact match, fuzzy match or perform a natural human translation.

Hello Igor,


I agree with you in all points, as I said I think CafeTran's take on the matter is more useful and efficient, at least for freelance translators.


It might be slightly inconvenient if an agency or project manager needs to prepare a project. They should instead provide the necessary (TMX) resources for the translator to perform the matching intelligently, on a segment per segment basis. 


The question on interoperability was off-topic and hadn't much to do with pretranslation, but with CafeTran's own states and if they are preserved if one needs to work on CT XLIFFs outside of CT.


The latest changes open some new possibilities, and some points I've made discuss just that, 

Login to post a comment