Saturday, December 12, 2009
L’informazione traduttiva necessaria
This post is the conclusion of the two preceding ones, which should be read first.
One of the sections of Alexander Ludskanov’s magnum opus that Bruno Osimo has preserved in his Italian recension is, fortunately, 3.2.1 L’informazione traduttiva necessaria (The Information Needed for Translating). It was the subject of my last discussion with A.L. in Ottawa not long before he died.
I dwell on it here because it’s a matter that’s fundamental to all translating, whether machine, expert or natural. Only a layman thinks that a good knowledge of the two languages involved is enough information. For example, even if you know the verbs, the pronouns, the sentence forms, you can’t even translate appropriately such a simple message as “How are you?” into French, Spanish, etc., without also being informed what the social relationship is between the speaker and the hearer: Comment allez-vous? or Comment vas-tu? The term A.L. uses for such information is extralinguistic.
A.L. was acutely aware that the acquisition and incorporation of extralinguistic information presented a major problem for machine translation (MT). A few years before, in 1962, the Israeli logician and linguist Yehoshua Bar-Hillel (see photo), who was engaged in machine translation research at MIT, had declared it to be an insurmountable obstacle to the realization of what he called Fully Automatic High Quality Machine Translation. His pessimism put a damper on MT research in the United States. One of his examples was the apparently simple phrase slow neutrons and protons. In order to translate it correctly into languages that require agreement between nouns and their adjectives, it must be parsed either as (slow neutrons) + protons or as slow (neutrons + protons). But the choice between them depends on prior knowledge of, or newly acquired information about, nuclear physics. To A.L., as a semiotician, the extralinguistic information did not come directly from the world outside the translator (the ‘real world’) but from what was coded in other sign systems in the translator’s mind.
On the other hand, it’s not necessary, for the translation of a given text, that the translator possess, or have access to, the whole vast ocean of human knowledge. Each translation requires only a few drops from the ocean, and that, perhaps, might be acquirable and could be processed by a computer. Generally speaking, the more narrowly specialized the source text is, the less of the ocean is needed. But how to determine and specify precisely what information is required? That’s what we saw as the primary problem of l’informazione traduttiva necessaria.
Reference (see also the two preceding posts)
Yehosua Bar-Hillel (1915-1975). The future of Machine Translation. Times Literary Supplement, April 20, 1962. Bar-Hillel called extralinguistic information encyclopedic information. He later recanted in part and admitted that perhaps MT didn’t need to be High Quality or Fully Automatic. There’s an article on him in Wikipedia, which also tells us that his granddaughter, Gili Bar-Hillel, is the Hebrew translator of the Harry Potter series.