Notes on analysis of automatic translations of "Michel Petrucciani n'est plus"
The automatic translations offered by Google (G) and RerversoPro (RP) illustrate the different levels of difficulty contained within the text. In general, translation software has a restricted dictionary and very limited syntactic parsing rules. The more idiomatic, the more figurative a word or expression, particularly the latter, the greater the difficulty faced by the software. Thus expressions like "le gratin du jazz" are understandably beyond the capabilities of automatic translation software. Even further removed from the scope of automatic translation is cultural allusion; for example, "la république" referring to the French Republic. Following are some of the more notable problems that students have commented on:
The dictionaries do not contain proper nouns. Thus "Le Devoir" is treated as two words, the second a common noun = "(the) Duty". Thus "châteauneuf-du-pape" is treated as three words, the first of which is not in the dictionary and therefore is not translated. Thus "Orange" is treated as a common noun, which just happens to be the same in English = "Orange".
More annoying is the software's inability, though understandable, to deal with the syntactic complexities of personal pronouns and personal adjectives. F il = E he or it; F elle = E she or it. F Son/sa/ses agree in gender with the possessed, whereas E his/her/its agree with the possessor.
Articles, definite, indefinite and partitive. French and English function quite differently in regard to the use of articles. Another complex area of comparative grammar where automatic translation software is powerless. Many examples in the text, of which apposition is a characteristic case: "Virtuoise du piano, Petrucciani..." = E "A virtuoso...".
Order of modifier and modified (adjective + noun, adverb + verb/noun, etc.). Another extremely complex area of syntax. One example: "Django Reinhart d'abord, Stéphane Grappelli ensuite, Michel Petrucciani enfin".
Polysemy. E.g. "Atteint... (d'une grave maladie)" = G "Reached...", RP "Reached{*affected*}...". RP at leasts indicates the existence of polysemy, leaving it up to the human translator to choose the correct equivalent. Example 2: "Tiens, (Michel)" = literal (verb tenir) vs. idiomatic (interjection).
Following are the two cases that would seem to present the most glaring errors in the automatic translations:
"Quand mon fils en joue" = G "When my son out of cheek", RP "When my son in cheek". This demonstrates the total absence of any syntactic parsing. Each program has interpreted "en" as a preposition and "joue" as a noun. (The mistake would perhaps be less severe if "en joue" had been translated as "take aim".) There is formal and functional confusion at the level of the separate words: en = préposition/pronoun, joue = noun/verb.
The prize for the real perle of the automatic translation should probably go to "la passion du père que les fils partagèrent très tôt" = G "the passion of the father whom the wire divided very early" = RP "the passion of the father which the threads shared very early". If RP avoided the "whom" and "divided" of G, both made a complete mess of "fils", which the most basic dictionary should give as the plural both of fil and of fils. This is a case of homonymy.
|