loge-gh / jp-tools Goto Github PK

Warodai has examples of two types:
- Expressions, ex. 気になる, i.e. stuff which should be converted to EDICT 
articles
- Actual examples, ex. 彼を殺した時に私は安心した (doesn't actualy 
have this line), which should eventially go to Tanaka corpus format file.

To separate these two, the only option I can think of now is to look the 
expression in English EDICT.

Some templates, such as ~suru, should not be applied to the word, but replaced 
with appropriate grammar tags, such as suru-verb. This is the convention used 
by JMDict.

Perhaps we are also to check all expressions against the same set of rules 
because if there's only one template it could be precompiled, i.e.

> SOUZOUsuru .... translation

We're to notice that SOUZOUsuru matches ~suru and convert it appropriately.

There's a lot of informal markers in the articles:

> кн. рыбы.
> спорт. судья; рефери (в борьбе сумо́).
> бот. спора.
> энт. тли, Aphididae.

> ономат.:
> 1) со стуком (падать, ударяться и т. п.);
> 2) жадно, давясь (пить, есть);
> 3) одиноко, потерянно.

We're to find these and in each case determine if the marker can be safely 
removed and added as appropriate grammar/topic tags.
If not, we're to ban the article in the strict mode.

JMDict has <s_inf> to store comments about a sense usage etc. Use this instead 
of putting this information into glosses.

This is a problem of Wakan dictionary format, where it merges together all 
grammar tags from a list of entries.

But even when this is fixed in Wakan (by converting to EDICT2 style 
dictionaries), we'll have no way of knowing, for the same kana-kanji pair which 
translation entry on EDICT relates to which translation entry on Warodai.

So we'll have to skip the cases where grammar flags are not for the whole 
article.

At most we can try to find common flags, the ones repeating in every entry, and 
apply those as common for translation.

Ex.:

第一[の] [だいいち[の]] /(adv,n) первый; перен. 
наилучший/(P)/
第一[に] [だいいち[に]] /(adv,n) во-первых; прежде 
всего; первым делом/(P)/

This confuses EDICT parsers and they treat [の] as a reading.

Articles sometimes have source language information (examples are not 
exhaustive):

> (фр. adieu) до свидания!, прощай[те]!

> (ит. alto)
> 1) альт, контральто (голос);
> 2) альт (инструмент).

> (кит. цзяоцза) пельмени.

We're to extract it and store in JMDict's <lsource> tag and in appropriate 
EDICT/2 forms:

> <sense>
> <lsource xml:lang="dut">ontembaar</lsource>