Re: [Ltru] RE: For review: Tagging text with no language

On 4/13/2007 9:24 AM, Mark Davis wrote:
> I always like to think of these kinds of issues by looking at 
> examples, since it tends to focus the issues and make it clear when 
> people are misinterpreting others' terminology. I put out below some 
> examples of what a process should do if gets a stream of information 
> and is to tag it, where we assume that it is doing the best job it 
> can. People can comment on these or propose others.
>
> Content
>  Tag
>  Comment
> n/a
>  und, or equivalently
> "" , if that is available in the protocol  The tag where the process 
> is not equipped to analyze the text at all. und = "Undetermined"
> 143kl;ufa)iop(&uweiorqhjkl2341lkj#@!$Jkdfj;afe  zxx
>  Clearly some binary junk. zxx = "No linguistic content"
> bok23
>  und
>  Maybe has linguistic content, maybe not. Can't really determine.
> chat  mul, if the protocol only permits a single tag
> <en, fr> otherwise  mul = "Multiple languages"
> maybe also others, since "chat" has entered the vocabulary of many 
> languages
> Suzuki  ja-Latn
>  maybe also others, since "I bought a Suzuki" is a perfectly 
> reasonable English sentence.
> Igonda flatunicai vbinkli?  mis  some language the process recognizes, 
> but which is not in BCP 47
> podstatné jméno  mis
>  something the process recognizes as having linguistic content, and 
> might be in BCP 47, but it doesn't know which language it is.
> if (myInstance.getType() == Type.UNKNOWN) { throw new Exception(""); }
>  art?
>  unclear whether "art" can include, or is restricted to cases like 
> Klingon or Esperanto. art = "Artificial (Other)"
>
Your suzuki example would benefit from context.

The "Suzuki" in "I bought a Suzuki" is clearly a proper name which 
doesn't change the fact that the entire text is in some form of English, 
while "Suzuki" appearing in context of Japanese text would indeed be ja-Ltn.

I sent out, a while ago, a list of possible edge cases (abstract, not 
concrete examples). You might look a them to see whether any others from 
that list should be given examples.

A./

Received on Friday, 13 April 2007 22:25:50 UTC