lexibank / bodtkhobwa Goto Github PK

The data is a good example for cldf in extended versions, and it contains:

• morpheme boundaries, to which morpheme tags are attached (all checked)
• partial cognates
• partial alignments

So we should be able to see here how well lexibank can account for these enhanced ways of annotation. And also if we can compute extract these things and convert them to cldf without major efforts.

The REPL spits a number of warnings, e.g.

[WARNING] Problem with row 821 in col 1, expected «<function at 0x7f7279777488>» as datatype but received «249 ? 250» (ROW: pʰ yː + tɕʰ u - + l u ŋ|249 ? 250|cockroach|cockroach|249 685 250|Khoitam|ʨʰu+luŋ|INSECT1 COCKROACH1 COCKROACH2|||570 685 250|i n + i n + i n c|pʰ yː + tɕʰ u + l u ŋ|(pʰyː) ʨʰu-luŋ, entry cogids).

during the makecldf call. This should ideally be addressed before the SDATA submission to make the reproducible run 'more clean'.

With the cross-semantic cognate sets, we can now make rather precise suggestions as to which "tags" we should give to morphemes. So the initial table needs to be refined, and we list only the 900+ morphemes, and we can in fact automatically already define how they should be called, if the occur only for one concept. This reduces the work greatly.

It'll be important to

• write generic code for this (up to @LinguList)
• note the steps in our worfklow for later paper on this or blogpost (@monpasang)
• adjust the howto shell script (@LinguList)

@monpasang should also find about half an hour (more not needed) to explain to @MacyL, what he was doing so far, so she can understand our workflow better, for future analyses of her own.

Concepts need to be double-checked, also with new concepts we want to add to concepticon, currently with an asterisk *.

As you will see from this pattern, final glottal stop release on vowels is often in fact corresponding with -k finals in other dialects. So re-writing this as a full glottal stop makes sense, as it allows us to capture final -k vs. final glottal stop as sound correspondence.

The raw file (see here) uses Ø to indicate that a value is missing. The question is here: is this value missing because there is no word for this entry, or is it missing, because it is not cognate with the rest?

We will assume that it is the latter for now, but it may be important, that Tim goes back and adds missing data.

I just created a context sensitive orthography profile. This allows you, @monpasang, to refine potentially problematic IPA characters.

It elicits data according to frequency. You best edit in LibreOffice or other spreadsheet software.

Many cases are unproblematic. In fact, what you have to do is: read the symbol in the first column as the one that we find in the data, and the second column as our preferred way of rendering it. Then you have examples and languages. If we add <?>, this means, we can't interpret the symbol. Most of the time, this means that something's wrong.

However, line 9 is our practice:

• we render morpheme boundaries with a + character, so you should put a + in there.

Note that $ means "end of word" (so only useful to know) and ^ means "start of word". We don't list them in the second column, as this is just information for us.

Note another question: Do you make a distinction between morpheme boundaries and word boundaries (e.g., by writing "-" and " " = space)? If so, I changed them spaces to _, as this is the way we render spaces in our system.

Diphthongs in this dataset are not recognized by CLTS:

The concept STICH should be spelled as STITCH.

It looks like a fork of concepticon/concepticon-data and a non-released commit is used here:

On Wikipedia (https://en.wikipedia.org/wiki/Kho-Bwa_languages), this CLDF dataset is referred to as:

Johann-Mattis List, Mei Shin Wu, Robert Forkel, & Tiago Tresoldi. (2019). lexibank/bodtkhobwa: Lexical Cognates in Western Kho-Bwa (Version v2.0) [Data set]. Zenodo. doi:10.5281/zenodo.3537604

This should be:

Timotheus A. Bodt, & Johann-Mattis List. (2019). CLDF dataset derived from Bodt's "Lexical Cognates in Western Kho-Bwa" from 2019 (Version v2.0) [Data set]. Zenodo. http://doi.org/10.5281/zenodo.3537604

This code has been written by @LinguList already for work with Burmish and N. Hill. It is in the dev/ part of the lingrex package and should be used for this. It should also be added to the util.py part of the lingrex package.

Here is how I envision the work

• First check of the data
  • convert to a preliminary wordlist for edictor
  • find missing words, so they can be added later (if needed)
  • make a preliminary orthography profile to see where we are with the data (and manually correct it)
• Work on the language list and link languages to glottolog (so we have their coordinates)
• link the concepts to concepticon (also to filter out cases where we have, e.g., "run1" and "run2"
• use the new concepts to make a wordlist for edictor that can then be manually edited
  • apply rudimentary partial cognate detection to it (@monpasang, maybe read our paper with Nathan Hill on this matter, so you understand)
  • refine the partial cognates
• work (computer-assisted) on the identification of language-internal cognates with edictor (maybe make one automatic investigation first), in order to identify cognates across meanings
• apply the correspondence pattern detection algorithm, and output the patterns, so this can be used for semi-automatic reconstruction
• make an initial version of the data for the lexibank repository, so @MacyL can see how we handle conversion to CLDF

In general, this is what I hope we can achieve in the week with @monpasang, but even if it's less, it will be a lot what we can do, as the data is in a good state.

please put the file with corrected cognates in the evening, before finishing, into the github, @MacyL will do this, and @monpasang will send the final file to here. Place it with a date-stamp to the folder work/, where we keep our annotations.

My name is Timotheus Adrianus Bodt. Not Timothaeus Adrianus Bodth.

See lexibank/pylexibank@2533f67#diff-7437d82378d1fce6944dd496b80307bf.

Should I have a look at the cross-concept cognates @LinguList or do you think that won't be necessary? There a couple of interesting ones: belly-belt-fruit, horse-frog, empty-head-knee-lungs, etc.

This CLDF dataset was also used to come to the following articles, which should be mentioned in the repository on Zenodo:

Bodt, Timotheus Adrianus (accepted, 2021). The Duhumbi perspective on Proto-Western Kho-Bwa onsets. Journal of Historical Linguistics.

Bodt, Timotheus Adrianus. 2019. The Duhumbi perspective on Proto-Western Kho-Bwa rhymes. Die Sprache 52 (2016 / 2017) 2: 141-176.

Ideally, we do this on Friday, have to prepare on Thursday. Once correspondence patterns are determined, we use the workflow of the work with Nathan to produce semi-automatic reconstructions for the data. Ideally we also have the tiers-check prepared, so we can make a first test here with @tresoldi (decision trees not needed yet).

If there are no glottocodes, add the coordinates. The filee is here, editing is also possible via github (just press the edit button). Tim Bodt should do this, once he's told me his github account.

I recommend to use the following link:

http://edictor.digling.org/?remote_dbase=khobwa.sqlite3&file=khobwa&basics=DOCULECT|CONCEPT|VALUE|FORM|TOKENS|STRUCTURE|MORPHEMES|CROSSIDS|ROOTIDS&root_formatter=CROSSIDS&highlight=TOKENS|ALIGNMENT&sampa=SEGMENTS#customize_anchor

or to copy the link from here

http://edictor.digling.org/?remote_dbase=khobwa.sqlite3&file=khobwa&basics=DOCULECT|CONCEPT|VALUE|FORM|TOKENS|STRUCTURE|MORPHEMES|CROSSIDS|ROOTIDS&root_formatter=CROSSIDS&highlight=TOKENS|ALIGNMENT&sampa=SEGMENTS#customize_anchor

and paste it into the browser (google wil also work)

@monpasang, please use this link, and we can even tweak it with more custom settings.

E.g.: nin ~ nun should be rendered as n i~u/i n in the TOKENS.

We just automatically linked the concepts of the file by Tim to concepticon. Now they need to be refined. Tim will work on this (approx. about 2 hours of work). The file is here, but this time, it is recommended to work on spreadhseet editor and not online.

we need to define clear-cut syllable structure templates, they also help to identify unsegmented multi-morpheme words.

forgot to do that before

@monpasang spotted some problems (misspellings), which should be handled. He'll add them in issues.

This work was funded by

Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung: Strategische Zielsetzungen im Subkontinent (100015_138331)

and

Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung: ‘Reconstruction of Proto-Western Kho-Bwa’ (Early Postdoc.Mobility-Stipendium Nr. P2BEP1_181779)

among others.