I remember when total recall was first released that Igor or someone said it was basically useless when translating Japanese or other language without spaces between words.
With it being so prominent in the new release, it just got me curious if this was still the case. I have not tried to create a file or try to use the feature based on previous feedback.
So, how is it?
I read your post. I tried to open the text file, but it was mojibake.
You seemed to indicate yourself that Total Recall was basically worthless.
What good are matches that barely reach the 50% threshold?
BTW, mojibake of Japanese characters can be avoided by changing the encoding when opening the resource (in many cases, to UTF-16).
OK, thanks for the clarification. However, I still don't understand what is useful about matches that barely reach a 50% threshold. Could you explain that?
The main goal of Total Recall is to avoid loading huge translation memories to RAM memory. Here, CafeTran recalls and loads to the working RAM memory only the segments that are needed for the current project context. The number of hits per word can be freely changed and the preliminary memory matching can still be used to increase the matching speed and program's responsiveness even further. You can sort of compare the Total Recall database to Long-term memory from which only the necessary data are retrieved into the Short-term memory.
IK: Here, CafeTran recalls and loads to the working RAM memory only the segments that are needed for the current project context.
All very well, but how does it [not] work for Japanese? I have an informant in the area at this very moment, and she'll report back to me, also if she'll have to apply Aikido.
If the source language has no word separator (e.g. Japanese and Chinese), CafeTran performs the analysis of the project on the character level. Then it is a good idea to increase the number of hits to 1000 as M suggested.
Igor, so does that mean it has value now?
I am asked about it back on the Google thing before this forum was developed, and I am pretty sure the consensus was basically that it wasn't worth setting up for languages like Japanese.
OK, well anyway, I decided to go ahead and create a table and then do the recall from memory set to 1,000.
I have left my main TM in there for comparison purposes.
After clicking through just a few different segments of my current project, I am getting hits in my actual TM, but just a nice blank screen in the total recall memory.
Yeah, definitely doing something wrong, the total recall memory ended up having 0 entries.
> OK, well anyway, I decided to go ahead and create a table and then do the recall from memory set to 1,000.
Did you actually import any TMX memories to Total Recall after creating a table?
Well, yes, of course. Two, actually.
A bit strange, right?
Please do as follows:
1. Activate Total Recall menu via View > Show Total Recall menu.
2. Select your created table via the menu Total Recall > Memory tables >
Can you see the table tab with any segments?
Now, go to the menu Total Recall > Show memory table info... and check the name of the database in the title bar of the info panel. It should say SQLite 3.8.7
Yes, there are segments.
The title bar says "XXX in H2 1.4.185 database"