SemanticImport mangles Greek letters in column headers

Question

Bug persisting through 13.1.0

---

I have a Dataset whose column headers contain Greek letters. If I export the Dataset as a csv file and then SemanticImport that file, the Greek letters appear incorrectly:

    data = Dataset[{ 
            <|"\[Epsilon]" -> 1.5, "\[Eta]" -> 1.0, "\[Nu]u" -> 5.3, "\[Mu]u" -> 3.23, "x0" -> 3.0 |>,
            <|"\[Epsilon]" -> 2.5, "\[Eta]" -> 2.0, "\[Nu]u" -> 5.4, "\[Mu]u" -> 15.3, "x0" -> 4.1 |>
        }];
    Export["test.csv", data, "csv", CharacterEncoding -> "UTF8"];
    data2 = SemanticImport["test.csv", CharacterEncoding -> "UTF8"]

The result is:

I can open the csv file with Excel to see that the Greek letters have been exported correctly. They are also imported correctly into Mathematica if I use Import instead of SemanticImport. (I know that I can convert the result of Import into a dataset as a workaround.)

I'm using Mathematica 13.1 on MacOS 11.6 (BigSur).

Answer 1

It looks like SemanticImport simply ignores the CharacterEncoding option and imports as "String" instead. Compare with

Import["test.csv", "String"]

"ϵ","η","νu","μu","x0"

1.5,1.,5.3,3.23,3.

2.5,2.,5.4,15.3,4.1

The same is true for SemanticImportString:

SemanticImportString[Import["test.csv", "Text"]]

It think it is a bug in SemanticImport, please report it to the official support.

As a workaround, use Import:

Import["test.csv", "Dataset", "HeaderLines" -> 1]