Trying to get libmecab.dll (MeCab) to work with C#

Question

I'm trying to use the Japanese morphological analyzer MeCab in a C# program (Visual Studio 2010 Express, Windows 7), and something's going wrong with the encoding. If my input (pasted into a textbox) is this:

一方、広義の「ネコ」は、ネコ類（ネコ科動物）の一部、あるいはその全ての獣を指す包括的名称を指す。

Then my output (in another textbox) looks like this:

?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
(   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
)   åè©ž,サ変接続,*,*,*,*,*
?   åè©ž,サ変接続,*,*,*,*,*
?????????????????????????   åè©ž,サ変接続,*,*,*,*,*
EOS

I would guess that that's text in some other encoding being mistaken for UTF-8-encoded text. But assuming that it's EUC-JP and using Encoding.Convert to turn it into UTF-8 doesn't change the output; assuming that it's Shift-JIS and doing the same gives different gibberish. Also, while it's definitely processing the text - that's how MeCab output is supposed to be formatted - it doesn't appear to be interpreting the input as UTF-8, either. If it were doing so, there wouldn't be all those identical lines in the output starting with one-character "compounds," which it's clearly unable to identify.

I get yet another different-looking set of gibberish when I run the sentence through MeCab's command line. But, again, it's just a row of single question marks and parentheses going down the left, so it's not just the problem that the Windows command line doesn't support fonts with Japanese characters; again, it's just not reading the input in as UTF-8. (I did install MeCab in UTF-8 mode.)

The relevant parts of the code look like this:

[DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl)]
private extern static IntPtr mecab_new2(string arg);
[DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl)]
[return: MarshalAs(UnmanagedType.AnsiBStr)]
private extern static string mecab_sparse_tostr(IntPtr m, string str);
[DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl)]
private extern static void mecab_destroy(IntPtr m);

private string meCabParse(string jpnText)
{
    IntPtr mecab = mecab_new2("");
    string parsedText = mecab_sparse_tostr(mecab, jpnText);

    mecab_destroy(mecab);
    return parsedText;
}

(In terms of fiddling with plausible-looking things to see if they make a difference, I've tried switching "UnmanagedType.AnsiBStr" to "UnmanagedType.BStr," which gives the error "AccessViolationException was unhandled," and adding "CharSet=CharSet.Unicode" to the DllImport parameters, which turned the output into just "EOS".)

This is how I've been doing the conversion:

// 65001 = UTF-8 codepage, 20932 = EUC-JP codepage
private string convertEncoding(string sourceString, int sourceCodepage, int targetCodepage)
{
    Encoding sourceEncoding = Encoding.GetEncoding(sourceCodepage); 
    Encoding targetEncoding = Encoding.GetEncoding(targetCodepage);

    // convert source string into byte array
    byte[] sourceBytes = sourceEncoding.GetBytes(sourceString);

    // convert those bytes into target encoding
    byte[] targetBytes = Encoding.Convert(sourceEncoding, targetEncoding, sourceBytes);

    // byte array to char array
    char[] targetChars = new char[targetEncoding.GetCharCount(targetBytes, 0, targetBytes.Length)];

    //char array to targt-encoded string
    targetEncoding.GetChars(targetBytes, 0, targetBytes.Length, targetChars, 0);
    string targetString = new string(targetChars);

    return targetString;
}

private string meCabParse(string jpnText)
{
    // convert the text from the string from UTF-8 to EUC-JP
    jpnText = convertEncoding(jpnText, 65001, 20932);

    IntPtr mecab = mecab_new2("");
    string parsedText = mecab_sparse_tostr(mecab, jpnText);

    // annnd convert back to UTF-8
    parsedText = convertEncoding(parsedText, 20932, 65001);

    mecab_destroy(mecab);
}

Suggestions/taunts?

Answer 1

I came across this thread looking for a way to do the same. I used your code as a starting point and this blog post for figuring out how to marshal UTF8 strings.

The following code gives me properly encoded output:

public class Mecab
{
    [DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl, CharSet=CharSet.Unicode)]
    private extern static IntPtr mecab_new2(string arg);
    [DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl, CharSet = CharSet.Unicode)]
    private extern static IntPtr mecab_sparse_tostr(IntPtr m, byte[] str);
    [DllImport("libmecab.dll", CallingConvention = CallingConvention.Cdecl, CharSet = CharSet.Unicode)]
    private extern static void mecab_destroy(IntPtr m);

    public static String Parse(String input)
    {
        IntPtr mecab = mecab_new2("");
        IntPtr nativeStr = mecab_sparse_tostr(mecab, Encoding.UTF8.GetBytes(input));
        int size = nativeArraySize(nativeStr) - 1;
        byte[] data = new byte[size];
        Marshal.Copy(nativeStr, data, 0, size);

        mecab_destroy(mecab);

        return Encoding.UTF8.GetString(data);
    }

    private static int nativeArraySize(IntPtr ptr)
    {
        int size = 0;
        while (Marshal.ReadByte(ptr, size) > 0)
            size++;

        return size;
    }
}