This post is a comparative test of Greek-to-English Automatic Translation in various sites, using one Greek sentence as test-data. Some of the results turned out to be… rather amusing. E.g.
Original Greek Sentence:
«Το ερώτημα είναι πως θα καταργήσουμε τα σύνορα, χωρίς να παραβλέπουμε την πραγματικότητα και χωρίς να πέφτουμε στην παγίδα του εθνικισμού.»
«Robotic Pigeon-English» translations:
1) InterTran (click here to test):
- «Το quiz they are πως θα καταργήσουμε the frontiers , devoid of to παραβλέπουμε την reality and devoid of to πέφτουμε στην trap of the nationalism».
2) Babel Fish Altavista (click here to test):
- «The question is that we will suppress the borders, without we overlook the reality and without we fall in the trap of nationalism».
3) Google translation (click here to test):
- «The question is how to do away with borders, not to mention of the fact and without falls into the trap of nationalism».
The results show that automatic translation software is not only inadequate, but also suffering from inexcusable forms of ignorance:
1) Intertran translation was proved to be -by far- the worst (useless, in fact)! The Greek Natural Language processing used by it has serious lexical inadequacies as well as grave grammatical errors, possibly due to incomplete Greek grammar rules and bad Greek syntax definitions.
2) Babel Fish / Altavista scored very badly. However, (at least) it tried harder than Intertran and it was proved to be ciompletely free of Intertran’s serious lexical inadequacies (at least on a surface level).
3) Google translation was the best: It translated correctly the entire first part of the sentence (before the comma) but it did blunders after that point. The Greek phrase «χωρίς να παραβλέπουμε την πραγματικότητα» (correct translation=»Without overlooking reality») was wrongly translated «not to mention of the fact». This is evidence that Google’s «Sense Disambiguation code» sometimes… tries harder than necessary : -The simpler translation «not overlook» (of «μην παραβλέπουμε» ) was… overlooked (amusingly enough) adopting something more sophisticated semantically: «not to mention» – which was… pure nonsense! (whereas Google’s competitor Altavista didn’t… try so hard, so got THIS one right)
Finally, the last part of Google’s translation, «falls into the trap of nationalism.» indicates that the agent or subject of the verb «falls» is missing from the entire translated sentence: Apparently, Google translation does little or no «post-processing» of the results, leaving serious semantic gaps open!
- UPDATE: A friend told me (on the phone) that maybe this particular mistake was not so big, after all, since Google translation might have meant «falls» in the PLURAL (as in «The Niagara Falls»). Ah well, it still makes little sense, when you look at the sentence as a whole. And I am not really sure if it was meant in this way.
Ah well, all
these big software companies typically employ conventional programmers
, who use procedural
programing languages like PHP and C++. They very rarely if ever advertise for hiring PROLOG
To a Prolog programmer (especially a veteran like myself!) all these problems can be solved (or at least reduced drastically) in various ways (a bit too technical to describe here).
I mean -for Heaven’s sake- the Greek test-sentence was NOT that difficult! Come on…
Well, most big software companies (and ALL Greek software companies) do not and cannot do «such things», since they are «not interested in hiring Prolog programmers» . However, soon enough this might prove to be their worst competitive disadvantage:
- Some of us (PROLOG programmers) are already working on the idea! 😉