Hi Jed, Google have published some details of their algorithm and that's pretty much how it works.
If they want to do say English/Italian translation they find a lot of text (books, menus etc.) that exist in both languages and then they analyse the text counting words by frequency. This gives first mapping between words. There's a lot more to it but frequency mapping is a key element. They must have trained their system using some books where Rossi had been translated to Smith. It's interesting that the computer learns to translate just by analysis of these dual language texts and with very little human input or language understanding. Colin On Fri, Dec 23, 2011 at 4:29 AM, Jed Rothwell <[email protected]> wrote: > Giovanni Santostasi wrote: > > Rossi doesn't mean Smith. It is translated sometime by Google as Smith >> because Smith is such common name in the anglophone world and Rossi is an >> extremely common (if not the most common) Italian last names. >> Rossi means "red one", probably the ancestors of this family were red >> headed. >> > > Wow! That gives us an interesting look at how Google translation works. > The computer picks a word that is functionally similar. One that has > similar uses, distribution or frequency. > > Or maybe it is a database error. > > The word "roth" also means red, in Middle English. Hence the placename and > family name "Rothwell" means red well. That is, a well with reddish water > from iron minerals in the water. See: > > http://www.rothwelltown.co.uk/**historyofrothwel.html<http://www.rothwelltown.co.uk/historyofrothwel.html> > > - Jed > >
