Some of the values in the Year column of values.csv are not years:

ID=1043: 'via lexicon'
ID=1757: 'ngarinyin1970'
ID=2162: 'yupik1990'
ID=2244: ''
ID=2247: 'rembarrnga1975'
ID=2503: 'soo1971'
ID=2560: 'wa2012'
ID=2882: 'tlingit1917'
ID=2928: ''
ID=2932: ''

• Korana: Work from Maingard is based on data from 1879 - correct in data, but not in Table 2 of the data
• Reference 1 for Nganasan has the wrong year in the Table; should be 1854; again, this is correctly coded in the data, but not in the table
• Reference 2 for NYangi has no year in the Table, but is correct in data

I had those cases marked in the production check, but they did not get changed.

Just came across this dataset. Looks cool.
One suggestion, though: Since the whole point of the dataset seems to be highlighting language change over time, wouldn't it be more transparent to list doculects as rows in LanguageTable, rather than Glottolog languages? This would get across the point that languages at different points in time might need to be treated as different more thoroughly - while aggregation on Glottolog language level would still be possible via the glottocode property.

In parameters.csv, the column names are like "Variable_type" vs. "Category_Esp" vs. "Description_esp". In the Python code, there were used different terms, I had to modify them. So I suggest to unify spelling (capitals, not capitals, etc.).

Would it be possible to give me push-access for this repository? Then I could adapt some of the changes that we still need to make before publishing the database.

The link from README.md to
https://github.com/cldf-datasets/gata/blob/main/cldf/Generic-metadata.json
is broken.
Not sure, why this might have happened - maybe you just need to recreate the README.

In the spreadsheet in raw/pooled_all.csv, you list sources in a colloquial style, not as their bibtex keys. They should be parsed in such a way that we have:

BibtexKey[pagerange]

So something like:

Muller (1963:555)

should be converted to:

Muller1963[555]

But note that this requires that the bibtex-keys are AuthorYear.

The ValueTable (values.csv) has multiple values coded as "NA" which is due to R being used somewhere in the pipeline. It would be better to code these as "" or tag these values as null in the StructureDataset definition (where ? and <empty string> are already defined).

I need admin rights to transfer the directory :)

@MuffinLinwist, what is the status on teh cldf conversion, are there any errors now?

To have us work further on this, please transfer to https://github.com/cldf-datasets, we then keep working from there.

The repository names we use are typically lower case, this is easier to handle. So I suggest to change the name to gata.

It will be better if we add codes to the dataset.

@LinguList The journal asked us to provide the data in one of their featured repositories. The most reasonable options would be Zenodo and OSF. As many other CLDF datasets are also linked to Zenodo, I guess that this would be the most reasonable way. Does anything speak against creating a release 0.1 with the current form of the dataset? We would like to not delay the submission any further, and while there are some open issues (#13 for some utility plots, #10 codes that could be added for a Version 1.0, and #5 which is too much work to implement it right now for little reward), we think that they should not prevent us from creating this release for submission.

If you are okay with that, I would go ahead and create a release 0.1 and link it to a Zenodo page. Alternatively, we would put everything on an OSF repository for the submission process.

@LinguList, I made some modifications on the dataset but still run with three problems:

1. It seems that the link with parameters.csv is not working because values.csv doesn't show this column. The same happens with languages.csv.
2. There is a problem with the Source column (I get cells with something like this as an output: a;r;a;p;a;h;o;1;9;6;3). I don't know why this occurs.
3. I tried deleting the Code_ID column. Everytime I run cldfbench makecldf, however, it appears again.

Currently, all plots are created using relative paths to languages and parameters. We can use rcldf instead and make use of the utility there.

See: https://github.com/SimonGreenhill/rcldf