Morphology guessing: it’s got a stem stuck in its head!

Question

ме

… is a prefix in our language. It never occurs as a stem. But the morphology guessing has got it into its head that it’s a stem, even though there are scores of words where I’ve morphologised it as a prefix.

ме is a prefix on around half of verb forms, and the guessed morphologies all have it as the stem, meaning that none of them are correct.

Today I decided to put some effort into this, so I sorted the list by stem, and corrected the few places where a morphology with “ме” as a stem had been approved by another user. The guesses didn’t change.

Then I ticked Show Revised Words Which No Longer Exist In Project^†, and found one more word with a “ме” stem. But with that corrected, the guesses with “ме” stems still didn’t change.

Is there a solution?

^† BTW, this sentence looks strange to me with initial capitals. I was taught at school that in titles, insignificant words like “in” should not be capitalised. Personally, I think that options like this should be in sentence case, with just the first word capitalised.

较早的帖子 - 以原始语言显示

Answer 1

I’d love some help on this one if anyone’s in the know!

较早的帖子 - 以原始语言显示

Comments

I’ve looked at the code and don’t see anywhere that we remove lexical
entries, so my guess is that your Lexicon.xml file for the project still
has a stem entry for “me”.

Search this file for: Form=“me”

If you find it and would like to try to remove it, you would have to delete
the whole … element that contains the text that you found.

Not sure of all the consequences of doing this, so I would recommend trying
it on a restored backup of your project first.

We probably need a way for those using the interlinear to see the
dictionary that Paratext is using and make modifications to it. One problem
with doing this is that we start doing more dictionary work that is already
available in FLEx.

John+Wickberg
Paratext Support

较早的帖子 - 以原始语言显示

---

John+Wickberg:

Search this file for: Form=“me”

Form="ме" occurs twice. For the benefit of others, this search gets to precisely the stem:

<Lexeme Type="Stem" Form="ме"

I discovered that ме was in the text as a word – in a regex I’d saved in book XXF. I must have noticed this before, because I’d morphologised ме as /xxxx/, presumably in the vain hope of stopping it matching other places where ме is a prefix.

So I did what you said, and created a scratch copy of the project. I deleted the little section beginning with the above string. The morphology of ме on its own didn’t change. I closed Paratext. It did not resave Lexicon.xml, but on opening PT, I got:

Is there something else I need to change in the file so that PT thinks its integrity has been preserved?

较早的帖子 - 以原始语言显示

---

I wrote to soon – I’d deleted the first line of the next section. Now I’ve corrected that, there’s no error, and my guess morphologies look great !

I have a question though: PT’s methods are supposed to be statistical. I had one ме in the Wordlist as a stem, but I’d parsed it as a prefix maybe 100 times. So why was PT still thinking it was more likely to be a stem?

较早的帖子 - 以原始语言显示

---

The real statistical part of the interlinear is in the matching of words in
the model to words in the text being worked on. This code finds the most
likely matches between the two texts by comparing the distribution of the
words across all the verses they occur in.

The creating of analyses is not statistical, but just finds all possible
parses of a word.

John+Wickberg

较早的帖子 - 以原始语言显示