ABSTRACT
The principal assumption behind machine interpreting is precisely that interpreting is a sequence-to-sequence, conduit model problem. A decent human interpreter would be aware of the social requirements of the situation and would automatically pick the right pronoun to use; a machine is stuck with whatever is in its database. In MT, the presence of words on a screen makes it possible to do all kinds of correction and editing between the computer producing its best guess and the text reaching a reader. First dreamt up by computer scientist Donald Michie in 1961 when computing itself was still in its infancy, the machine learning approach to noughts-and-crosses ironically doesn’t need any computers at all. Another way to programme a computer to play noughts-and-crosses would be simply to feed it all the possible games of noughts-and-crosses, all 255,168 of them, and then instruct it to run statistical analyses on them to find patterns.
