Order of suggested glosses in interlinearizer tool

Question

What is the order of the suggested glosses in the interlinearizer tool?

In other words, if you click a gloss and get the drop-down list, is the order just the order in which the various possible glosses happened to be chosen over the history of the project? It doesn’t look like it’s ordered by frequency, though that seems like it would be most logical.

Comments

This may require a new thread, but perhaps this thread would be the proper place.

One thing that frustrates me with the Interlinearizer is when I add a new word parse, Interlinearizer seems to go through the entire corpus and assign that word parse as the new suggestion for each time this wordform occurs, overwriting my previously confirmed word parses (1st line, which populates the 2nd line). As my work continues then, increasingly rare word parses overwrite all the common ones (in the 2nd line), so I always need to go through the Interlinearizer with a particular eye for detail before showing it to anyone outside the specific language team who doesn't know the language. For example, we've decided to write contractions in speech (contracting /a e/ '3s pres' to /'e/), but when I did this the Interlinearizer changed every accepted /e/ wordform to the suggestion /a e/ & I have to change them all back (still finding them every once in a while, when we've gone back to revise some earlier work). This also happens when I create a new word parse for a compound word. The word in question and any homonyms now get a word parse suggestion second line if one wasn't there previously. For example, when I added the word parse /ba den/ (mother and child) for the word /baden/ 'sibling', every other /ba/ (big, ocean, or) was assigned the wordform suggestion 'mother' too. I have missed a few of such changes when showing our work to consultants, and it is confusing to them, not to mention embarrassing for me.

So, my question is whether there is a way to avoid this happening or if it needs to be a feature request?

---

Reading through your question, I'm not exactly sure what you're saying is happening.

Are you saying that when you choose a less common gloss, that all future *blue* suggestions give that gloss rather than the more common ones? That surprises me--I would have expected PT to either chose the most common gloss as it's suggestion, or to go with the most recent. In the 2nd case, I would have expected just go back to your preferred option the next time you chose that one.

Or are you saying that creating a new, unusual gloss is overwriting your already approved glosses? If so, that's most definitely a bug.

---

I was a little hasty on this. It does not overwrite already approved glosses. I double-checked that. However, it does seem to put the the inappropriate word parse (second line) when there was not a separate word parse. So, the example again is /e/, which I'd only entered on the bottom line as 'Present Tense'. When I found one place where /e/ was a contraction for /a/ '3s' plus /e/ 'Present Tense', that became the automatic word parse assigned for every other /e/ in the text. But it does seem to populate this line with the most recent word parse, because I suggested something else for this line (/a e a/, I think), and all the word parses in blue for /e/ afterwards were changed to that (/a e a/). I do not know, however, if this behavior would go away if I restarted the Interlinearizer because I deleted the word parse without restarting the Interlinearizer.

---

This is getting detailed enough that it might be necessary to have someone connect to your computer and visually see what you're describing. If you're glossing morpheme-level parts on one line and then the whole word on a second line, you're outside my area of knowledge.

One thing I wonder is what you mean by "delete". In the interlinearizer that would refer to two possible actions. One would be that you go to the blue/black highlighted word, click the gloss, and hit backspace. I can't imagine that that action would affect future predictions, but I'm not 100% sure about that.

But the second option is to click the gloss and then hit the X next to the gloss you don't like. That's dangerous because it deletes that gloss in every single case, even in places where you've approved it. But if you hit the X, I'm quite positive that that gloss would never be suggested as a prediction .

Answer 1

The glosses appear in the list in the order in which they are approved. Here is a brief video describing how the order can be changed: https://www.youtube.com/watch?v=d-1eC90zW5M&index=1&list=PL3e7fqHIximoODBRd1GMcfZZ4GpcoGoQ7

WARNING: this involves editing a file outside of Paratext. Paratext should be closed before you attempt this and great caution should be used any time you edit a file outside of Paratext.

Comments

Thank you. I had assumed that editing that file would cause the results the video shows, but doing that by hand seemed too time consuming (and subjective).

It would be nice if someone smarter than me could create a program–either within PT or external–which would edit that Lexicon.xml file based on frequency. I guess it would search the Interlinear_XX_XXX.xml files to extract frequency numbers.

Answer 2

So is there a way to remove guessed glosses when some of them are obviously wrong?

Answer 3

The way to remove the guessed glosses is to replace them with the real glosses. Paratext should then learn the correct glosses.